PUBLIC UNDERSTANDING OF SCIENCE IN AMERICA
A Thesis
Presented to the faculty of the Department of Public Policy and Administration
California State University, Sacramento
Submitted in partial satisfaction of
the requirements for the degree of
MASTER OF PUBLIC POLICY AND ADMINISTRATION
by
Scott Cantrell
SPRING
2012
© 2012
Scott Cantrell
ALL RIGHTS RESERVED
ii
PUBLIC UNDERSTANDING OF SCIENCE IN AMERICA
A Thesis
by
Scott Cantrell
Approved by:
__________________________________, Committee Chair
Robert Wassmer, Ph.D.
__________________________________, Second Reader
Su Jin Gatlin Jez, Ph.D.
Date
iii
Student: Scott Cantrell
I certify that this student has met the requirements for format contained in the University
format manual, and that this thesis is suitable for shelving in the Library and credit is to
be awarded for the thesis.
, Graduate Coordinator
Robert Wassmer, Ph.D.
Date
Department of Public Policy and Administration
iv
Abstract
of
PUBLIC UNDERSTANDING OF SCIENCE IN AMERICA
by
Scott Cantrell
The United States is losing its global dominance in the fields of science and
engineering largely because of underinvestment in research and development and the
shift toward more knowledge-intensive industries that have greater emphasis on
intellectual capital, science, and technology. Critical thinking skills, complex reasoning,
and writing skills are lacking in undergraduates and recent college graduates in the
United States. Scientific literacy is necessary for the U.S. to maintain its global
competiveness and to also enable citizens to participate as more informed members in a
pluralistic society. National survey data on scientific literacy in America collected by the
National Science Foundation and General Social Survey indicates a deficit in factual
scientific knowledge and understanding by the American public. National cross-sectional
data from the 2008 General Social Survey was used in an Ordinary Least Squares
regression to see if socioeconomic, political, and religious factors were significantly
associated with scientific literacy, defined as the percentage of correct responses to 26
science and technology questions. Statistically significant regression coefficients were
found on independent variables representing education level, generational cohort,
income, gender, ethnicity, religiosity, and population size. Graduate, bachelor, and
v
associate degrees had the largest positive association with scientific literacy. Black,
Hispanic, and American Indian ethnic groups had the largest negative association with
scientific literacy. A gender gap was also found with females showing lower scientific
literacy. Policy interventions to assist ethnic groups and female K-12 students to stay in
school, earn their high school diplomas, and improve their readiness for college are
warranted.
, Committee Chair
Robert W. Wassmer, Ph.D.
Date
vi
TABLE OF CONTENTS
Page
Chapter
1. INTRODUCTION .........................................................................................................1
Purpose of Thesis .....................................................................................................1
Science and Politics .................................................................................................4
Evolution and Creationism ......................................................................................6
Climate Change Denialism ......................................................................................8
Science, Technology, and Education .....................................................................10
Science and the Media: Sources of Information ....................................................13
Organization of Thesis ...........................................................................................14
2. LITERATURE REVIEW ............................................................................................16
Socioeconomic Factors ..........................................................................................18
Political Context.....................................................................................................21
Religion ..................................................................................................................24
Literature Review Conclusions ..............................................................................26
3. METHODOLOGY ......................................................................................................29
Introduction ............................................................................................................29
Analytical Approach ..............................................................................................29
Source of Data........................................................................................................32
Model Specification and Broad Casual Factors .....................................................34
vii
Independent Variables ...........................................................................................37
Alternative Specification of the Model and Interaction Variables ........................46
Multicollinearity ....................................................................................................46
Heteroskedasticity ..................................................................................................48
Regression Coefficients and Ninety Percent Confidence Interval .........................49
Conclusion .............................................................................................................50
4. RESULTS ....................................................................................................................51
Introduction ............................................................................................................51
Descriptive Statistics and Correlation Matrix ........................................................51
Regression Model Analysis ...................................................................................53
Interaction Variables ..............................................................................................61
Multicollinearity and Heteroskedasticity ...............................................................62
Regression Coefficients and Ninety Percent Confidence Intervals .......................62
Overall Fit of the Model ........................................................................................67
5. CONCLUSIONS AND RECOMMENDATIONS ......................................................70
Appendix A. Science and Technology Questions..............................................................89
Appendix B. Data Tables ...................................................................................................95
References ........................................................................................................................114
viii
LIST OF TABLES
Tables
Page
1.
Dependent and Independent Variables .....................................................................95
2.
Descriptive Statistics .................................................................................................98
3.
Correlation Matrix ..................................................................................................100
4.
Quadratic Model Regression Coefficients ..............................................................108
5.
Ninety Percent Confidence Intervals for Statistically Significant
Coefficients .............................................................................................................111
6.
Worst-case and Best-case Scenario ........................................................................113
ix
1
Chapter 1
INTRODUCTION
Purpose of Thesis
Since the mid-1950s, social scientists in the United States and Great Britain have
been collecting national survey data that measures the public’s understanding of science
and technology (Miller, 2004). There is an entire field of study focused on scientific
literacy and its implications for public policy 1. The journal Public Understanding of
Science was first published in 1992 and includes articles based on both qualitative and
quantitative research (Bennett & Jennings, 2011). Scientific literacy generally refers to
the public’s factual knowledge of science and the scientific process, including its overall
objectives and general limitations (Miller, 2004; National Science Foundation, 2010). It
also means having the ability to critically read and evaluate information and assess its
veracity. Miller (2004) suggests scientific literacy requires one to have a basic
understanding of the vocabulary used to describe scientific terms and concepts as well as
a general understanding of the process used in scientific inquiry. He also suggested the
public should have sufficient understanding of science to be able to read and comprehend
the science column of The New York Times, The Wall Street Journal, or other major
newspapers and magazines. Miller (1998, 2004) calls this level of scientific knowledge
and understanding “civic scientific literacy.”
1
The term public understanding of science is preferred in Britain whereas in the United States the
terms scientific literacy and public knowledge of science are used in the literature more
commonly. I will use all these terms interchangeably in the thesis.
2
Civic scientific literacy is essential in virtually all aspects of American life to the
public as members of the workforce, consumers of science and technology, and citizens
of the body politic (National Science Foundation, 2010). Having a basic understanding
of science can improve our knowledge of ourselves and give us a sense of our
relationship to the natural world. It can help us evaluate the risks associated with taking
certain medications or undergoing specialized medical procedures. It can help us
understand the public debate on evolution and whether Creationism should be taught in
high school science classes. It can help us weigh the risks we face with climate change
and provide us a basis on which to evaluate governmental policies to reduce greenhouse
gas emissions linked to global warming. In a democratic form of government, it is
essential its citizens make informed decisions on public policy issues and know how to
discern facts from opinions in political discourse.
The purpose of this thesis is to investigate the socioeconomic factors associated
with the public’s understanding of science in the United States. The most basic question
I seek to answer is what is the contribution of political ideology, religiosity,
demographics (age, gender, marital status, income, ethnicity, and geographic region of
the country), and education to an individual’s understanding of science? Knowing these
contributions will help policymakers craft approaches and programs to address the gap in
scientific knowledge and understanding found in the American public.
In the remainder of this chapter, I discuss why the thesis topic is an important
public policy issue. I give examples of the politicizing of science; the religious right’s
3
push to teach Creationism in public schools; climate change denialism; how America is
falling behind other nations in areas of education, science, and technology; and sources of
scientific information in the media and why they cannot be wholly relied upon. Each of
these examples illustrates the high stakes involved when the public lacks a basic
understanding of science and critical reasoning abilities. I describe how conservative
politics can undermine science itself through a variety of means, including the creation of
pseudoscientific disciplines and scientific-sounding arguments; the suppression of
science by political appointees; and by magnifying scientific uncertainties to create
controversy where none really exists. Religious faith can also act as a perceptual filter
that screens out factual, scientific information.
I also illuminate recent national educational assessment data on grade school and
high school students and a recent longitudinal study of university students showing high
numbers of students are coming out of these educational institutions with only very basic
levels of scientific understanding and critical reasoning abilities. The United States is
losing its competitive advantage in science and technology against other nations,
especially China, because of its underinvestment in research and development. Finally, I
depict data showing most Americans obtain information on science through the Internet
and television. Given the vast amount of information available through these media,
some valid and some spurious, it is critical consumers of this information have critical
reasoning abilities.
4
Science and Politics
Several books have been published recently describing how scientific illiteracy is
posing great threats to scientific progress, the environment, public health, and our
nation’s standing in the world with respect to education and economic competition. The
author of Denialism (Specter, 2009) argues that illogical thinking, irrational fear of
change, and political and religious ideology are hindering scientific progress and threaten
our future. The author describes the public’s mistrust of scientists and cynicism toward
corporations, including pharmaceutical companies and the healthcare industry, to explain
an unusual group of organized interests opposing vaccinations. These interest groups
have established web sites with government-sounding names, including the National
Vaccine Information Center 2 and websites less subtle in concealing their advocacy
mission, including Vaccination Liberation. 3
Mooney (2005), author of the New York Times bestseller titled The Republican
War on Science, describes the political misuse of science by social conservatives, which
pinnacled during the George W. Bush Administration from 2001 to 2009. During this
period, there was a sharp increase in the number of former oil and coal lobbyists that
became political appointees to key federal government agency positions under the Bush
administration. Mooney provides evidence to show that President G.W. Bush was openly
anti-science and allowed ideology and politics to trump scientific consensus. Even
though G.W. Bush made a campaign pledge in 2000 to regulate carbon dioxide
2
3
http://www.nvic.org/
http://www.vaclib.org/
5
emissions, one of his first actions in office was his refusal in 2001 to sign the
international agreement enforcing the Kyoto Protocol to reduce greenhouse gas emissions
linked to global warming, citing concerns over cost, loss of jobs in the United States and
questioning the validity of science. 4 He also endorsed teaching “intelligent design” (a
rebranding of creationism) in high school biology classes.
Mooney (2005) describes the rise of the modern American conservative
movement, which has increasingly dominated the Republican Party’s agenda over the last
50 years. He describes key players such as William F. Buckley, Jr., American author and
commentator, who founded the rightwing journal National Review in the mid-1950s,
which had a major impact on the conservative movement. Mooney (2005) then describes
Arizona Republican senator Barry Goldwater’s right-wing anti-intellectualism and
distrust of the educated elite and the impact he had on the scientific community. Mooney
(2005) also discusses the impact of President Nixon, George H. W. Bush and George W.
Bush on scientific institutions and policies within their respective administrations
culminating in the modern American conservative movement, which is a political
philosophy that strongly resists changes to existing social and institutional structures and
opposes “big government” regulation of industry and business. It is also a moralistic
philosophy that has co-opted the Religious Right to expand its sphere of political
influence.
4
http://usliberals.about.com/od/environmentalconcerns/p/KyotoProtocol.htm
6
As a recent example of how religion has crept into politics, the Sacramento Bee
reported (Eckolm, 2011) that four potential candidates for the 2012 Presidential election
including Mike Huckabee, Newt Gingrich, Haley Barbour, and Michele Bachman,
founder of the House Tea Party Caucus, spoke at a two-day conference for pastors run by
the Pastors Policy organization on March 24-25, 2011 in West Des Moines, Iowa. 5 The
Policy Pastors organization has conducted conventions in at least 14 states in recent
years. At these conventions, pastors have been told of the secular threats to conservative
Christian values such as the sanctity of marriage, the rights of the unborn child, and the
Divine right of freedom from big government. Pastors have been told it was their
Christian duty to speak out against the excessive reach of the U.S. Environmental
Protection Agency and the secular liberals attempt to take religion out of politics.
Mooney (2005) believes it is critical for more scientists to speak out against
pseudoscience and bogus claims made by science-abusing conservatives to preserve the
integrity of the scientific process and the future of our planet. One of the best examples
of pseudoscience involves the creation of scientific-sounding arguments and institutions
to promote a religious belief and refute evolution. This subject is discussed in the
following section.
Evolution and Creationism
In the General Social Survey (GSS) conducted by the University of Chicago in
2008, only slightly more than half (50.5%) of respondents agreed with the statement that
5
As of March 2012, Gingrich, Romney, Santorum, and Paul are the main contenders for the
Republican Party nomination as a candidate to challenge President Obama in 2012.
7
human beings as we know them today evolved from earlier species of animals (General
Social Survey [GSS], 2010). 6 It has been 150 years since the publication of Charles
Darwin’s seminal works on the origin of species, and currently only about half of the
general public accepts evolution even while biologists are documenting evolutionary
changes and adaptation in some species on a time scale of 10-20 years (Pitchford, 2010).
Despite a preponderance of scientific evidence from various disciplines that the universe
began with an event called the Big Bang 14 billion years ago, that the earth is 4.5 billion
years old, and that humans evolved from earlier forms of primates, the public is still in
denial about these well-established scientific theories. Many persons instead choose to
cling to religious, faith-based convictions that are in denial of these facts. Young-Earth
creationists, for example, believe the earth was created exactly as described in the Book
of Genesis less than 10,000 years ago and that the Great Flood reshaped it. Since religion
cannot be proved in fact and is instead based upon “faith,” when facts counteract faith,
the religious must usually choose faith or they would lose their religious beliefs.
Evolutionary theory contradicts dogmatic religious beliefs and literal
interpretation of the Bible as the inerrant Word of God. Religious faith does not require
skills in critical reasoning or standards of evidence nor peer review and hypothesis
testing. It only requires faith and belief. Creationists have assumed the garb of science
by creating pseudoscientific disciplines like Creation Science, Intelligent Design Theory,
and Global Flood geology and establishing organizations like the Institute for Creation
6
http://www3.norc.org/GSS+Website/Download/SPSS+Format/
8
Research (2012) and the Intelligent Design and Evolution Awareness Center, a 501(c)(3)
non-profit organization promoting intelligent design theory, particularly focusing on
university and high school students. 7 This is particularly concerning because these
religious activities are targeting students and undermining science curricula which may in
turn lead to students having a misinformed understanding of the nature of science as well
as lack the ability to plan and conduct scientific investigations.
Climate Change Denialism
The cover story in the August 2007 edition of Newsweek magazine was on
climate change denialists and the strategies being employed by well-organized, industryfunded think tanks and scientists who have created a miasma of doubt and uncertainty
about climate change. Climate change denialists use a variety of tactics such as paid
advertisements in the media, opinion editorials in major newspapers, white papers,
studies, and lobbying to create uncertainty around the science of climate change. Climate
change science has been called fraudulent, a great hoax, junk science, and a liberal
environmentalist conspiracy.
In fall 2009, a computer hacker stole 1,000 e-mails and 3,000 documents from the
Climatic Research Unit at the University of East Anglia in Britain. These e-mails spoke
of scientists massaging and falsifying data, hiding data showing declines in global
temperature, deleting incriminating e-mails, and so forth. Conservatives seized this
opportunity to proclaim that global climate change is nothing but junk science based on
7
http://www.ideacenter.org/
9
lies; some of the more extreme bloggers asserted that “globalists” are part of a liberal
conspiracy to establish a world government. Since 2009, the global scientific community
studying climate change has been in the mode of defending its research and deflecting a
barrage of scornful criticism from the right-wing media and blogosphere. Unfortunately,
it is usually the loudest and most persistent messages that get most of the attention in the
media and that in turn shape public opinion.
In response to the controversy that arose over these e-mails stolen from the
Climatic Research Unit, Senator Inhofe (Oklahoma-R) called for an investigation by the
inspector general of the Commerce Department of the National Oceanic and Atmospheric
Administration (NOAA) scientists implicated in the e-mails (Kaufman, 2011). In a
February 2011 report by the Obama administration, the inspector general stated that no
evidence of scientific misconduct could be found or that NOAA scientists inappropriately
manipulated data. NOAA’s top official was also exonerated of any wrongdoing.
In May 2010, a group of 255 members of the National Academy of Sciences
published an open letter in the journal Science, which was also sent to the White House
Office of Science and Technology. 8 The letter described grave concerns over the
“political assaults on scientists in general and on climate scientists in particular” (Sills,
2010, p. 689). The letter described threats they had received of criminal investigations by
Congress over e- mails between climate scientists leaked to the public. The authors of
the letter emphatically stated, “There is compelling, comprehensive and consistent
8
The National Academy of Sciences consists of about 2,100 members and 350 associates from
other countries (http://www.nasonline.org/site/PageServer?pagename=MEMBERS_Main).
10
objective evidence that humans are changing the climate in ways that threaten our
societies and the ecosystem on which we depend” (Sills, 2010, p. 689). And yet, despite
overwhelming scientific evidence of climate change, a large proportion of the public still
does not believe in global warming or that it is being largely caused by human activities.
In a 2010 survey conducted by the Yale Project on Climate Change
Communication, 38% of respondents said there was a lot of disagreement among
scientists whether global warming was actually occurring while 39% believed most
scientists think global warming is real (Leiserowitz, Smith, & Marlon, 2010). The survey
also showed that 63% of Americans believe global warming is occurring but did not
know why it was occurring. Nineteen percent of respondents believed it is not occurring
and the remainder said they did not know either way. The climate change issue is a good
example of the various tactics used to politicize science by undermining science itself,
targeting individual scientists, and magnifying scientific uncertainties (Gore, 2011;
Mooney, 2005).
Science, Technology, and Education
The National Assessment of Educational Progress (NAEP) provides a nationwide,
ongoing assessment of American students’ knowledge of mathematics and science,
reading and writing, arts, civics, economics, geography, and American history (National
Center for Education Statistics [NCES], 2011). The last assessment was conducted in
2009; however, recent changes in the assessment methodology mean that these results
cannot be compared to assessments done in prior years. Nevertheless, the 2009
11
assessment provides a benchmark for future assessments and presents interesting data on
America’s fourth-, eighth- and twelfth-grade students.
Achievement levels for the NAEP are set for each subject area and grade level
and are reported as the percentage of students performing at Basic, Proficient, and
Advanced levels. Only 1% of fourth graders and 12th graders and 2% of eighth graders
performed at the Advanced level in NAEP science. The remaining students performed at
either a Basic level (between 60% and 72%) or a Proficient level (between 21% and
34%) of science achievement. Another interesting finding of the 2009 NAEP report is
that student performance varied by geographic region, ethnicity, family income, and type
and location of school (public versus private school and city versus another location).
The San Francisco Chronicle (Tucker, 2011) reported that results from the NAEP survey
worried science education leaders because of students’ poor performance on scientific
subjects.
Another recent study paints a stark assessment of the critical thinking abilities of
college graduates, which may put our nation at a competitive disadvantage with other
nations. The McClatchy Washington Bureau (Rimer, 2011) describes a longitudinal
study of several thousand undergraduates from 24 colleges and universities through four
years of college. This study was conducted by New York University sociologist Richard
Arum, who found that 45% of these undergraduates lacked critical thinking, complex
reasoning, and writing skills after the second year of schooling. At the end of four years,
36% showed no significant improvements in these skills. It also showed that students
12
pursuing traditional liberal arts degrees in subjects such as humanities, social science,
mathematics, and science made greater gains in critical reasoning skills compared to
students in business, education, social work, and communications. An additional finding
of Arum’s study is that large numbers of students avoided taking classes requiring
substantial amounts of reading and writing (defined as more than 40 pages of reading a
week and more than 20 pages of writing in the semester). This study suggests a deficit in
critical reasoning abilities among college students that may be symptomatic of a broader
problem in our educational system in which critical reasoning skills are not being
adequately taught to students. These reasoning skills are also required for an
understanding of science and its methods of enquiry.
The National Science Board, which oversees the National Science Foundation and
the Foundation’s Division of Science Resources Statistics, produces a bi-annual report on
science, engineering, and technology data in the United States and foreign countries. The
Science and Engineering Indicators 2010 report (National Science Foundation, 2010)
describes trends in public knowledge, understanding, and attitudes about science and
technology (National Science Foundation, 2010). This report shows the United States is
losing its global dominance in science and engineering largely because of
underinvestment in research and development and the rapid growth of East Asian
economies. The global economy is also shifting toward more knowledge-intensive
industries that have greater emphasis on intellectual capital, science, and technology.
According to an article published in the BBC News (Shukman, 2011) on a new study
13
conducted by the Royal Society, China is projected to overtake the United States in terms
of the number of science publications in reputable international journals in just two years.
China has been investing heavily in research and development, growing by 20% since
1999, and now spends greater than $100 billion each year. It is also producing large
numbers of science and engineering graduates from Chinese universities. 9
Science and the Media: Sources of Information
In the GSS 2008 survey, respondents said their likely sources of information on
science were the Internet (53.8%), television (22.0%), books and other printed material
(8.0%), newspapers (5.9%), and magazines (5.0%) (GSS, 2010). Because of the
tremendous amount of both valid scientific information and spurious information
available to the public, especially through the Internet and TV, it is extremely important
for the public to have a basic understanding of science and its methods and have the
ability to distinguish pseudoscience from science and opinion from evidence-based
statements. However, providing the public with more scientific facts and information
may not be sufficient to improve scientific literacy because of the filtering effects of
ideology. Nisbet and Goidel (2007) developed a theoretical framework that describes the
“impact of value predispositions, schema, political knowledge and forms of mass media
use in shaping public perceptions of science” (p. 421). Their findings indicate that
religious and ideological values appear to act as perceptual filters of scientific
information.
9
In 2006, there were over 1.5 million graduates in the fields of science and technology.
14
Mooney and Kirshenbaum (2009) in the book Unscientific America suggest we
cannot lay the blame of scientific illiteracy entirely upon members of the public and that
scientists themselves could do a better job of communicating scientific information to the
public and media. Scientists traditionally have not seen public communication as their
responsibility and there are few incentives for doing so. The authors also describe the
dumbing down of information and the decrease of science and technology coverage in the
media over the last couple of decades. The media also does not always get its facts
straight because of either inadvertence or by design. Political ideology strongly
influences what gets covered in and the veracity of information reported by the media.
Political and religious ideology also acts as a lens through which the public sees the
world and shapes public opinion.
Organization of Thesis
In the next chapter, I discuss regression-based studies that evaluated the
contributions of age, gender, marital status, income, education, ethnicity, geography,
political ideology, and religiosity to scientific literacy. In Chapter 3, I describe the source
of my data and rationale for my choice of the dependent variable and its relevance to the
research topic as well as the choice of the independent variables and the functional form
of my model. In the results chapter, I describe the outcomes of my regression analysis
and make policy recommendations based on my findings. Understanding the relative
contributions of socioeconomic factors to scientific literacy can help us develop targeted,
15
evidence-based policies to increase public understanding of science in America and
increase our global competiveness in science and technology.
16
Chapter 2
LITERATURE REVIEW
Substantial research has been conducted in the United States and in Europe since
the mid-20th century to assess the various factors related to the public’s understanding of
science and technology, or what is also called scientific literacy. This topic has been
studied by eminent institutions, including the National Science Foundation, National
Opinion Research Center, which administers the General Social Survey (GSS) 10, British
Royal Society, and the European Commission. Researchers have investigated the many
dimensions of scientific literacy including the use of survey instruments to assess
respondents’ factual knowledge of science and its methods and to collect data on
socioeconomics, trust, attitudes about scientific research, and the value of science to
society, political affiliation, and religious beliefs (Sturgis & Allum, 2004).
Scientific literacy has been studied using survey instruments that include a battery
of questions related to science and technology and the general principles and methods of
science. Some authors have questioned the validity and reliability of these types of
“know what” surveys aimed at assessing simple textbook knowledge of respondents
without considering and controlling for cultural, social, and ethnographic factors to
explain differences in public understanding of science (Pardo & Calvo, 2004; Sturgis &
Allum, 2004). Smith (1996) discusses a concern that respondents may be more likely to
guess true than false on these types of surveys, which would inflate the number of correct
10
The GSS 2008 survey data was used for this thesis.
17
responses to questions that are in fact true. Guterbock participated in a workshop
sponsored by National Science Foundation examining the survey instrumentation and
adequacy of the questions used by the National Science Board to evaluate public
knowledge of science (Guterbock, 2011). He writes in the scientific proceedings:
It must be said at the outset that the current questionnaire items that form the basis
for the 2010 Science Indicators report rest on a solid social science foundation.
These indicators of “scientific literacy” have proven durable and serviceable in
practice, and have been widely applied both nationally and internationally. It is
clear that they pick up important, theoretically predictable variations in science
knowledge. (p. 4)
The questionnaire items Guterbock (2011) identified were included in the GSS 2008
survey. I believe the GSS provides useful data that can be analyzed to gauge public
knowledge of science in America.
I organize my review of the literature into three broad themes: socioeconomic
factors; political context; and religion. These are the primary themes I use to categorize
the independent variables I believe explain differences in public knowledge of science. I
expand on these themes in the next chapter of my thesis, which describes the
methodology and statistical model I used to perform the analysis.
I believe socioeconomic factors, including age, gender, marital status, income,
education, ethnicity, and where one lives in America, may explain differences in
scientific knowledge and understanding. In the political context, I believe political party
18
affiliation and ideology are also explanatory factors. Finally, I think religion and views
on the Bible, whether one has dogmatic religious beliefs or no religious affiliation at all,
will help explain variation in scientific knowledge and understanding. These factors are
explained in detail in Chapter 3. I now justify these beliefs through an examination of
how these causal factors have appeared in the previous literature on this topic.
Socioeconomic Factors
Smith (1996) wrote a report on scientific knowledge around the world for the
National Opinion Research Center (NORC) at the University of Chicago. Based on
results from earlier studies of the public’s understanding of science, the author
hypothesized that scientific and environmental knowledge would be higher among men,
younger adults, higher educated people, the less religious, people with higher incomes,
people working in scientific fields including science teachers, and people living in major
urban areas. The author used data from the 1993 International Social Survey Program,
which administered surveys to 21 countries, to collect data on the public’s scientific and
environmental knowledge. One of these surveys was aimed at measuring scientific and
environmental knowledge and consisted of a 12-item questionnaire spanning the topics of
radiation, medicine and disease, atmospheric conditions, natural history, and astrology.
Response options consisted of Definitely True, Probably True, Probably Not True,
Definitely Not True, and Can’t Choose and were scored numerically from 1 to 5. Thus, a
perfect score would be equal to 12 and the worst possible score would equal 60 points. A
19
score of 36 meant that the respondent did not attempt to answer any of the questions
(Smith 1996). 11
The author found strong support for all of his hypotheses. Bivariate correlations
between socioeconomic factors (e.g., education, income, etc.) and environmental and
scientific knowledge had standardized coefficients 12 that were almost all statistically
significant in the hypothesized direction and were found in almost all the countries
surveyed. The author found that education (one of the independent variables) had the
strongest association with scientific knowledge (the dependent variable), with
standardized coefficient averaging -.308 at the national level. The author found that
nations with higher levels of economic development, measured by per capita gross
national product, had increased scientific knowledge, possibly due to stronger science
education standards in the classroom and broader coverage of science in the media with a
standardized coefficient of -.217. This means that for each unit increase in income
(measured by per capita gross national product), scientific knowledge (as measured by
the 12-item survey) increases by .217.
The author also performed a multiple regression using the same variables
included in the national surveys. At the individual level, higher levels of education have
11
Smith (1996, p. 15) explains that because of the scoring system he used, a lower score
represents relatively higher scientific knowledge. Negative coefficients would be expected with
variables such as education, income level, and so on.
12
Standardization of variables is done when the units of measurement of each variable is different
so the variances are equal to one. Standardized partial regression coefficients (þi) tell us the
change in the dependent variable for each standard deviation change in the independent variable
(ρi). A high absolute value of þi indicates a strong influence on the dependent variable. The
statistical test of a partial regression coefficient is H0: βi = 0 (Zar, 1984).
20
the highest association with scientific knowledge with a standardized coefficient of -.196.
Occupations involving science or teaching careers have standardized coefficients equal to
-.060 and -.065 for each respective occupational group. Individuals with some form of
religiosity, measured by church attendance, belief in God, Protestant and Catholic,
showed equivocal results regarding the effects on scientific knowledge.
The European Commission has been conducting public opinion surveys in its
member countries since 1973. It monitors such topics as social context, health, culture,
economics, politics, and education. A survey instrument called the “Eurobarometer” was
administered by the European Commission in 1992 in 12 European countries to assess
scientific literacy and collect socioeconomic data in the European Union (EU). The
survey consists of a 12-item science quiz covering the scientific fields of biology,
geology, physics, and chemistry taught in the curricula of primary and secondary schools.
Pardo and Calvo (2004) analyzed the formal qualities of the Eurobarometer test,
including the distribution of correct responses in the total sample (12 EU countries), with
the mean and variance of these responses. Differences in reliability were found due to
the structure of the test (e.g., relative proportion of true and false items, sequencing of
questions, etc.) and made recommendations on how to improve the test. They also found
variance in the discriminatory power of the test as a function of income level, years of
education, and a country’s level of industrialization.
Pardo and Calvo (2004) found it difficult to discern differences in more developed
countries and groups having higher education and income. They concluded that despite
21
some methodological issues with the Eurobarometer test, it was able to identify major
differences (not fine-grained differences) in scientific literacy in different societies and
social groups within EU member countries. For example, in a one-way analysis of
variance test for each nation, they found a highly significant positive relationship
between years of education and scientific literacy. Every country showed that individuals
with more than 20 years of education demonstrated a greater number of correct answers
on the test compared to those with less than 15 years of education (Pardo & Calvo, 2004).
The British Royal Society established an ad hoc group (“Working Group”) to
study ways in which public understanding of science could be improved and who should
be directly involved in these activities. The Working Group on the Public Understanding
of Science published its seminal report in 1985, which included strategic
recommendations to improve formal education, promote science in the mass media,
foster interest in science through museums and industry, and engage individual scientists
and the scientific community in public relations and outreach (Royal Society, 1985).
Political Context
Sturgis and Allum (2004) investigated how scientific knowledge (the independent
variable) influences public attitudes toward science (the dependent variable in their
model is “support for science”). They describe the deficit model, which says, in part, that
support for science is a function of the public’s understanding of science and what it
means to study something scientifically (Sturgis & Allum, 2004). The deficit model is
built on many years of survey-based tests of scientific knowledge. The authors also
22
describe a “contexualist perspective” that includes ethnographic information such as
cultural norms and public values in real-life settings to understand how these factors
influence the public’s understanding of science (Nisbet & Goidel, 2007, p. 422). The
authors constructed a model that includes all the variables in the deficit model as well as
an additional variable called “political knowledge” to represent the public’s knowledge of
governmental institutions and politics. The authors assume “political knowledge” can
serve as a proxy for how science is used to develop policies and regulations (Sturgis &
Allum, 2004, p. 58).
Sturgis and Allum (2004) found that scientific knowledge has a highly significant
positive coefficient (.254) which means that for every unit increase in scientific
knowledge (independent variable), support for science (dependent variable) increases by
about 25%. This result is consistent with deficit model assumptions that the effect of
scientific knowledge on attitudes toward science will be positive. They also tested the
contextualist model, which assumes the presence of “knowledge domains that influence
attitudes towards science and technology in opposite or conflicting ways to factual
scientific knowledge” (Sturgis & Allum, 2004, p. 58). The authors tested whether
support for science is influenced by political knowledge 13 and how the interaction of
political knowledge and scientific knowledge may affect support for science.
13
Political knowledge is based on the knowledge of policy positions of the main political parties
in Great Britain, measured by a six-item scale. This general level of political sophistication is
intended to be a general proxy for the respondent’s understanding of scientific institutions and
how they operate. The authors included an interaction variable called political x scientific
knowledge in the model.
23
Political knowledge had a highly significant positive coefficient of .087, but the
magnitude of the effect on science support was small. The interaction variable had a
larger positive coefficient (.158) demonstrating support for the contextualist model. The
top percentiles of both domains of knowledge increase favorable attitudes toward science
by nearly 30 percentage points. The Sturgis and Allum (2004) study is important because
it shows that increased scientific and political knowledge are determinants of support for
science.
Nisbet and Goidel (2007) developed a conceptual framework showing how value
predispositions, political knowledge, and various forms of mass media consumption
influence public attitudes toward controversial scientific and technical issues. For their
case studies, they conducted a nationwide random telephone survey to ascertain public
understanding and attitudes toward embryonic stem cell research and human cloning.
They found Christian conservatism had a negative coefficient of -.14, meaning that for
each one-unit increase in religiosity, holding constant all the other independent variables,
support for embryonic stem cell research decreased by 14%. With respect to support for
therapeutic cloning, they found negative relationships with social ideology (coefficient of
-.12) and age (coefficient of -.14), meaning that more conservative and older individuals
were less supportive of cloning. Level of education had a positive relationship with
support for cloning (coefficient of .13). The authors also found that regular media
consumption of Christian-based and science programming on television and time spent
reading religious and scientific newspaper and magazine articles were significant
24
explanatory variables. Reading newspapers and watching science fiction programs on
television had positive coefficients (.12 for both coefficients) that were highly significant,
suggesting that these media provide a substantial portion of the social context used by
individuals for evaluating these controversial science issues (Nisbet & Goidel, 2007).
Religion
Gibson (2004) states that public policy has traditionally been categorized into
developmental, allocational, and redistributional policies, thought to be driven by
socioeconomic factors and shaped by political subcultures. He argues for the inclusion of
a third category of policy he calls “morality policy,” consisting of core moral values of
individuals the author believes have strong influence on government policymakers. He
believes morality policies on issues such as abortion, gay rights, and the
creationism/evolution debate are becoming more forceful in shaping state moral values
and political outcomes, more so than even socioeconomic factors (Gibson, 2004).
Gibson describes political debate in America as a culture war between religious
conservatives and progressives and examined whether states with large proportions of
evangelical adherents had an impact on state morality policy, for which he used the
teaching of evolution in a state’s public school science curriculum as a proxy. 14 He
controlled for socioeconomic factors including state educational attainment level,
percentage of urbanized area in the state, and per capita income. The author used
14
The author cites a report by Thomas B. Fordham that evaluated state curricula with respect to
the inclusion of evolutionary theory in state science standards (Gibson, p. 1137). Each state was
given a letter grade A, B, C, D, or F based on how well it adhered to science standards.
25
multiple regression to test his hypothesis. His results show the coefficient of evangelical
is highly significant (-5.03) and is negatively associated with how well evolution is
included in a state’s science curriculum. He concluded that a state’s religious
environment is a good predictor of state policies on science standards, which supports the
morality policy framework.
Zigerell (2010) performed an OLS regression using data from the 2008 GSS. The
study used percentage correct responses to sixteen science and technology questions as
the dependent variable and found that respondents having a literal belief in the Bible had
less scientific knowledge as indicated by the negative coefficient (-.12) on the
independent variable called literal view of the Bible. As explained by the author, this
coefficient means that after controlling for gender, age, race and other factors,
respondents with a literal view of the Bible scored 12 percentage points less than the
excluded variable, which corresponded to just less than two questions out of 16 possible
(mean of 1.92 questions). The gap was still observed even after controlling for education
and demography; however, the difference was substantially lessened (coefficient of -.05).
Sustersic (n.d.) was interested in understanding how education and religion affect
belief in evolution. The author performed a logistic regression on the 2006 GSS dataset
to analyze key explanatory variables such as race, religion, and education (independent
variables) and how these factors affected the respondents’ beliefs in evolution (dependent
variable). The survey question was whether the respondent believed humans evolved
from earlier forms of animal life. Susteric’s (n.d.) paper was not clear on how regression
26
coefficients were estimated and determined to be significant; however, it was reported
that Black and Latino respondents had significant negative coefficients. These results
mean if a respondent was coded as Black or Latino, while controlling for all other factors,
the respondent would be less likely to agree with the statement that humans evolved from
earlier forms of animal life. The same negative relationship was found in fundamentalist
Christians and conservatives. Being a Catholic, believing that religion is important in
one’s life, and attending church service once or more a week, all had significant
coefficients but the direction of the effect was not uniform. Respondents who took high
school biology after 1963 was not found to be a significant factor associated with belief
in evolution. However, the variable “college graduate” had a substantial positive effect
on belief in evolution, while controlling for all other independent variables. The “postgraduate” variable had an even greater positive effect on the dependent variable.
Sustersic (n.d.) also included variables representing broad geographic regions of the
United States, but none were found significant.
Literature Review Conclusions
There are concerns about the validity and reliability of survey instruments (Pardo
& Calvo, 2004). It is not clear whether respondents are showing acquiescence bias,
which is a tendency to default to a “true” response for some questions. There is also
concern about the varying level of difficulty between the questions and how
representative these questions are of the public’s understanding of science. Researchers
have made recommendations on how to improve the survey methodology and suggested
27
different ways of measuring scientific literacy (Laugksch, 2000). I found it is difficult to
make comparisons or generalizations regarding the coefficients derived from the various
studies I reviewed because there were many different models being tested, at both an
individual and national level, along with differing assumptions and methodological
approaches by individual researchers.
Despite the caveats and limitations of the survey format used in GSS 2008, I
believe there is still much information that can be derived from this national survey
database using OLS regression methodology.
I was surprised to find so few articles having a direct bearing on my topic of
socioeconomic factors associated with scientific literacy in the United States. I found
only a few studies that used OLS regression. 15 At the individual level, it appears
scientific literacy is associated with level of age, education, income, ethnicity, religiosity,
and attitude toward science (Pardo & Calvo, 2004; Smith, 1996; Sturgis & Allum, 2004;
Sustersic, n.d.; Zigerell, 2010). There is also a positive relationship between scientific
knowledge and respondents with a favorable view of science that supports what is called
the deficit model 16 (Sturgis & Allum, 2004). At the national level there is evidence that
higher scientific knowledge is found in men, younger adults, higher educated people, the
less religious, people with higher incomes, people working in scientific fields including
15
In addition to my own database searches, I worked with two different research librarians on
separate occasions in the library at California State University, Sacramento to assist me in finding
relevant studies.
16
The deficit model is the concept that the public is deficient in its knowledge of science and,
therefore, doubts the value of science or is afraid of scientific innovation. This in turn leads to
unfavorable views of science and scientific institutions (Sturgis & Allum, 2004).
28
science teachers, and people living in major urban areas (Pardo and Calvo, 1996; Smith,
1996).
Evidence for the contexualist perspective exists, which says the interaction
between political knowledge and scientific knowledge can partially explain public
attitudes toward science (Sturgis & Allum, 2004). Value predispositions including
religiosity, political knowledge, and various forms of mass media consumption, all seem
to act as filters through which the public views controversial scientific and technical
issues, such as stem cell research and therapeutic cloning (Nisbet & Goidel, 2007).
Conservative religious views and ideology have been shown to be negatively
associated with a belief in evolution (Sustersic, n.d.). Persons with a literal interpretation
of the Bible have been shown to have less scientific knowledge compared to persons
having less dogmatic views; however, when controlling for demographic and educational
factors, the differences are greatly lessened (Zigerell, 2010). There is also evidence that a
state’s religious environment is a good predictor of science educational policies and
whether evolution is being incorporated into the science curriculum (Gibson, 2004).
I organized my discussion of the literature based on three broad themes:
socioeconomic factors, political context, and religion. In the next chapter, I expand on
each of these themes and describe variables that are proxies for each of these broad
categories.
29
Chapter 3
METHODOLOGY
Introduction
The purpose of this chapter is to describe the analytical methods and source of
data used in my investigation of differences in science knowledge among adults. I
describe how I measured knowledge variation across adults using 2008 survey data
collected in the General Social Survey. 17 I describe the analytical methods I used to
perform my analysis on these data and why they are justified, followed by the
specification of the functional form of my model. I then describe the broad categories of
factors that emerged as being important to explain variation in science knowledge found
in my literature review followed by a description of the dependent and independent
variables in my model. I conclude with a description of an alternative specification of the
model, my test of interaction variables, and how I tested for model bias. The remainder
of this chapter contains the following sections to achieve each of these purposes:
analytical approach; source of data; model specification and broad causal factors;
independent variables; alternative specification of the model and interaction variables;
multicollinearity; and heteroskedasticity.
Analytical Approach
My question is what is the contribution of political ideology, religiosity,
demographics (age, gender, marital status, income, race, ethnicity, and area of the
17
The GSS 2008 survey data was used for my thesis.
30
country), and education to an individual’s understanding of science? Institutions
including the National Science Foundation, the National Opinion Research Center
(NORC) which administers the General Social Survey (GSS), the British Royal Society,
and the European Commission have used survey instruments to help answer this
question. Researchers have investigated the many dimensions of scientific understanding
to assess respondents’ factual knowledge of science and its methods and looked for
associations with socioeconomic factors, public trust toward scientists and research
institutions, the value of science to society, political ideology, and religious beliefs
(Sturgis & Allum, 2004). Researchers have used both qualitative and quantitative
methods to study these various factors and their relationship with scientific
understanding. Regression analysis is a standard statistical method that has been used to
analyze survey data to determine the relative contributions of each of these factors to
public understanding of science (Bauer et al., 2007; Laugksch, 2000).
Regression analysis is a method that makes quantitative estimates of the
relationship between one variable, called the dependent variable, and one or more
independent variables. In simple regression, in which only two variables are considered,
it is a method of determining whether a functional dependence exists. We can never
prove causation between two variables, only correlation. But with a well-specified model
that controls for other factors expected to influence a dependent variable, we can be more
certain of causation. A regression model estimates the average unit change in the
31
dependent variable for each unit change in the independent variable and can be
represented as a linear equation:
Y = β0 + β1X + εi
where Y is the dependent variable, β0 is the constant or intercept term and β1 the
slope coefficient on the independent variable X indicating how much the dependent
variable changes with each unit increase of the X variable. These terms are called the
deterministic components of the equation. The term εi represents the random error
accounting for the variation in Y that cannot be explained or accounted for by the
deterministic components.
Functional dependence between two variables means the dependent variable in
some way depends on the value of the independent variable, whereas the reverse is not
true. For example, age may be a determinant of blood pressure, but blood pressure does
not determine age. There are also circumstances in which two variables may be
positively or negatively correlated with each other, but which lack any kind of functional
dependence. For example, daily high temperatures in Sacramento may be correlated with
daily ice cream sales, but sales of ice cream do not determine air temperature. In other
words, correlation does not imply causation. Proving causation requires more than just
reporting correlation coefficients; it requires a sound description of potential causal
mechanisms.
Multiple regression involves two or more independent variables and estimates
their effects on the dependent variable. Multiple regression analysis provides estimates
32
of how the value of the dependent variable changes with a change in any one of the
independent variables while all the other independent variables are held constant. The
partial effect of each independent variable is measured by the regression coefficient.
Ordinary Least Squares (OLS) regression is a standard method to calculate the regression
coefficients so as to minimize error of estimates in the model, or technically speaking, to
minimize the sum of the squared residuals (Studenmund, 2006). OLS regression is the
method I used to perform a cross-sectional analysis of GSS 2008 survey data (GSS,
2010). In my regression model, I examined whether there is functional dependence of
public knowledge of science on independent variables such as age, gender, income,
education, and political and religious ideology. I began my analysis with a linear model
but settled on a quadratic model as my functional form, which I explain in the next
section. I ruled out using a double-log and semilog functional form because the
dependent variable has five observations that include zero (i.e., the percent correct
responses were zero). In the next section, I describe the source of my data followed by a
description of my regression model.
Source of Data
The General Social Survey (GSS) is a national survey conducted since 1972 by
the National Opinion Research Center (NORC), part of the National Data Program for the
Social Science, headquartered at the University of Chicago and includes field offices
across the country. The GSS “[c]onducts basic scientific research on the structure and
development of American society with a data-collection program designed to both
33
monitor societal change within the United States and to compare the United States to
other nations” (GSS, 2010, para. 1). The GSS is partially funded by the National Science
Foundation and is one of NORC’s longest running projects.
The GSS conducts random, personal-interview surveys to collect core
socioeconomic and attitudinal data that lends itself to cross-sectional and trend analysis
and cross-national comparisons. The 2006 GSS included a constructed variable called
“science-quiz,” which was computed as the sum of the number of correct responses to 10
science questions (Pollack, 2009). For example, respondents were asked to answer “true”
or “false,” or “don’t know” to the following statements: electrons are smaller than atoms;
lasers work by focusing sound waves; and antibiotics kill viruses as well as bacteria.
The 2008 GSS included an additional 16 questions on science and mathematics
that were added to the 2006 survey to better reflect national standards for what K-12
students are expected to know (National Science Foundation, 2010). For example,
respondents were asked a question and given the option of several possible responses, but
there was only one correct answer. For example, respondents were asked questions such
as: Which of the following is a key factor that enables an airplane to lift? For which
reason many people experience shortness of breath more quickly at the top of a mountain
than along a seashore? What property of water is most important for living organisms?
These questions (and more) were included only in the 2008 survey and were not carried
forward into GSS 2010. Therefore, I decided to perform my analysis on the GSS 2008
dataset since it contained all 10 questions from the 2006 GSS plus the 16 science
34
questions added in 2008 (26 questions total). I performed all my statistical analyses
using PASW© Statistics 18. In the next section, I describe the functional form of my
statistical model, the broad causal factors that explain variations in public knowledge of
science, followed by a description of the variables that serve as proxies for these factors.
Model Specification and Broad Casual Factors
In this section, I describe the functional form of my model and how it relates to
my research question. In my literature review, I found studies showing scientific literacy
to be associated with socioeconomic factors including level of education, ethnicity,
gender, income, religiosity, and attitude toward science (Pardo & Calvo, 2004; Smith,
1996; Sturgis & Allum, 2004; Sustersic, n.d.). There is also evidence that higher levels
of political sophistication and scientific knowledge are associated with positive attitudes
toward science (Nisbet & Goidel, 2007; Sturgis & Allum, 2004). 18 Religious
conservatism is negatively associated with public knowledge of science and support for
scientific technology, including embryonic stem cell research and therapeutic cloning
(Nisbet & Goidel, 2007). I believe the broad categories of socioeconomics, politics, and
religion represents the range of potential factors that may be associated with public
knowledge of science.
18
I decided not to include variables in my model related to political sophistication and attitudes
toward science because I chose to focus on socioeconomic, political ideology, and religious
factors.
35
I constructed the dependent variable as the percentage of correct responses to 26
science questions in the GSS 2008 19. I believe the dependent variable (Y) is influenced
by the broad causal factors of socioeconomics, politics, and religion (represented by A,
B, and C). This relationship can be expressed algebraically as:
Y = f (A, B, C)
The broad categories on the right side of the equation above can be decomposed into
independent variables (x1, x2, x3 …xn), which serve as proxies for these broad categories
and can be expressed as:
A = f (X1, X2, Xi)
B = f (X4, X5, Xi)
C = f (X6, X7, Xi)
The functional form of my regression model can be specified as a linear equation 20:
Y = f (x1, x2, x3, … Xi) =>
Yi = β0 + βiXi + εi
where, i = 1, 2, …, n;
Yi = the ith observation of the dependent variable;
Xi = the ith observation of the independent variable;
εi = the ith observation of the random error term;
β0 = the intercept;
βi = the regression coefficient on the ith independent variable; and
n = number of observations
19
All questions are treated equally; there is no weighting of variables.
In the next section I describe how, through model iteration, I settled on the quadratic functional
form for my regression model.
20
36
The dependent variable, called “percentage of correct responses” in my regression
equation, is computed as the percentage of correct responses to 26 questions from the
GSS 2008 survey (see Appendix A). These questions were designed to assess the
respondents’ knowledge of the physical and biological sciences, understanding of charts
and statistics, reasoning ability, and use of experimental and controlling variables in a
scientific experiment. As stated in the National Science Foundation report (2010), “A
good understanding of basic scientific terms, concepts, and facts; an ability to
comprehend how science generates and assesses evidence; and a capacity to distinguish
science from pseudoscience are widely used indicators of scientific literacy” (pp. 7-17).
Factual questions have been included in science knowledge surveys by the National
Science Foundation, General Social Survey (GSS), British Royal Society, and the
European Commission for several decades. My approach, which was to use data derived
from performance on the 26 science questions in the 2008 GSS, is consistent with these
prior studies.
For each science question in the GSS 2008, there were several possible responses
but only one correct answer. I created dummy variables to represent the correct
responses for each of the questions. Ten of the questions required a response of “true,
false or don’t know.” Fourteen of the science questions were multiple-choice questions
for which there was only one correct answer. And finally, two of the questions required a
short narrative response for which there was also only one complete and correct answer.
37
Because there is only one correct response to each of the survey questions, it was simple
to compute the percentage of correct answers. 21
In the next section, I describe the independent variables (e.g., age, income,
gender, race, etc.) used in my model (see Table 1). I also describe the expected direction
of effect of each coefficient on the dependent variable. Descriptive statistics that include
the mean, standard deviation, and maximum and minimum values of each variable used
in the model are shown in Table 2. I used PASW© Statistics 18 to generate a table of
correlation coefficients between independent variables and to test for statistical
significance (Table 3). I checked for high correlations (> .8) between independent
variables that had significance at the .05 or .01 level (two-tailed test called Pearson’s r).
High correlations may be a sign of multicollinearity, which I discuss later in this chapter.
Independent Variables
Each of the broad causal factors shown on the right side of the equation below,
Percentage of correct responses = f (socioeconomic factors, political factors, religion
factors) + ε
can be decomposed into independent variables, which shall serve as proxies for each of
these broad categories:
Y = f (x1, x2, x3, … Xi) =>
Next, I describe each of these independent variables and their expected direction of effect
on the dependent variable.
21
I created dummy variables for correct responses and coded incorrect responses and “don’t
know” responses as zero.
38
Socioeconomic Factors
age = f {Baby Boomer (+), Generation X (+), Generation Y (+)}
The variable “age” is an interval-level variable coded in GSS 2008 as the exact
age of the respondent at the time of the survey. Zigerell (2010) used an OLS regression
and found a negative regression coefficient on age (-.14). I expect age may be an
explanatory factor of variation in scientific literacy due to generational differences of
respondents. I created dummy variables to represent persons born in the following
generational cohorts and show the expected direction of effect on the dependent variable
in parentheses: Traditionalist (the excluded variable) born 1900-1945 (-), Baby Boomer
born 1946-1964 (+), Generation X born1965-1980 (+), and Generation Y born 1981-1999
(+) (Lancaster & Stillman, 2002). 22
I expect members of Baby Boomer, Generation X, and Generation Y to answer a
higher percentage of science questions correctly compared to Traditionalists. Higher
scientific knowledge may be found in younger generations because persons in these
cohorts grew up at the time when personal computers started to become widely available
to households along with Internet access. Generation Y is a highly connected group of
individuals that has had nearly lifetime access to the Internet, including more recently,
Internet access on a wide variety of mobile devices. It is important to recognize the
Internet is the dominant source of science information (53.8%) followed by television
(22.0%), according to survey respondents (GSS, 2010). In addition, the creation of
22
Persons born since 2000 are in Generation Z, however; this cohort is younger than 18 and,
therefore, does not meet the minimum age requirement to participate in the GSS survey.
39
public, educational, and government access television programming since about 1970,
which includes channels devoted to science (such as NOVA, National Geographic,
Animal Planet, and Nature), has probably had a positive impact on public knowledge of
science overall.
gender = f {male (+)}
In the literature, I found only a few studies that looked at gender differences with
respect to performance on scientific questionnaires, but none of these studies reported
statistically significant differences (Miller, 1983; National Science Foundation, 2010;
Susteric, n.d.). Miller (1983) constructed an index of scientific literacy and found that
only 7% of respondents to a 1979 National Science Foundation survey were scientifically
literate. This subgroup consisted of primarily males, respondents over 35 years of age,
and college graduates. Miller (2004) used a structural equation analysis and found the
variable female had a negative direction of effect on civic scientific literacy (-0.24).
Zigerell (2010) used an OLS regression and found a negative regression coefficient on
female (-.08). The Science and Engineering Indicators 2010 report (National Science
Foundation, 2010) showed men scored higher than women on 15 of the 16 science factual
knowledge science questions added in 2008 as part of the 2008 GSS. The regression
study of belief in evolution by Susteric (n.d.) included a male dummy variable, but it was
not found to be significant.
I think there is some evidence that gender may have an effect on public
knowledge of science so I included a male dummy variable in my regression model. I
40
think there may be cultural stereotypes and expectations that women should not be as
good at math and science as men which may be reflected in the 2008 GSS performance
on science questions. This stereotype may gradually be changing as more women enter
the workforce and choose occupations with higher educational requirements.
marital status = f {married (+)}
I did not find any studies in the literature including marital status as a factor in
public understanding of science. However, I think it is plausible there could be a positive
effect on the dependent variable if the respondent is married and the spouse is college
educated. I included a married dummy variable (+) to test whether it had a significant
effect on scientific knowledge.
income = f {low income (+); moderate income (+), high income (+), and very
high income (+)}
I expected income to be a factor explaining some of the variation in public
knowledge of science, an idea also supported in the literature (Pardo & Calvo, 2004;
Smith, 1996). I expected that compared to very low income (the excluded variable),
higher levels of income would have a positive relationship with scientific literacy since
higher income affords greater opportunities to obtain information and to get a college
education.
Cook (2010) describes total family income changes from 2008 to 2009 based on
U.S. Census Bureau data from 2009 (U.S. Census Bureau, 2011), which categorizes total
41
family income into quintiles 23, which are fifths of the U.S. population ordered by their
amount of income. These quintiles break down as follows (Cook, 2010):
First or bottom income quintile: < $20,453
Second quintile: $20,454 to $38,550
Third quintile: $38,551 to $61,801
Fourth quintile: $61,802 to $100,000
Fifth or top income quintile: >$100,000
In GSS 2008 there are 25 categories of income ranging from under $1000 per year
at the low end to the highest reported income category of $150,000 or more per year.
Therefore, I could not use the specific income as an explanatory variable and had to
create dummy variables to represent income brackets. I took the GSS 2008 income data
and collapsed 25 categories into five income brackets, which are close approximations of
the quintiles described above. I then created low income (second quintile), moderate
income (third quintile), high income (fourth quintile) and very high income (fifth
quintile) dummy variables. The first quintile, very low income, is the excluded variable.
education = f {high school diploma (+); junior college degree (+); bachelor degree
(+); graduate degree (+)}
I expected educational attainment to have a positive direction of effect on
scientific literacy, which is also supported by the literature (Pardo & Calvo, 2004; Smith,
1996; Susteric, n.d.; Zigerell, 2010). I expected respondents having a high school
23
A quintile is one of the four values dividing a range of data into five equal parts consisting of
20% each.
42
diploma, junior college degree, bachelor degree, and graduate degree to have a positive
effect on scientific knowledge compared to respondents having less than a high school
education (the excluded variable). I created dummy variables for each of these categories
of educational attainment.
ethnicity24 = f {black (-); American Indian (-); Asian (-); Hispanic (-)}.
I found limited evidence in the literature suggesting race may be a factor in public
understanding of science. Whites, Blacks, and Hispanics comprised over 95% of the
sample while all other ethnic groups cumulatively made up the remaining percentage in
the GSS 2008 survey. These other ethnic groups consisted of Asian Indian, Chinese,
Filipino, Japanese, Korean, Vietnamese, some other Asian race, native Hawaiian, some
other Pacific Islander, or some other race. Cumulatively, these racial groups comprised
less than 4% of the sample. I decided to compress these categories and code them as one
group called Asian. I expected Hispanics, Blacks, Asians, and American Indian ethnic
groups to show a negative direction of effect on scientific literacy compared to Whites
(the excluded variable), potentially due to cultural differences. Susteric’s (n.d.)
regression analysis of GSS 2006 data to evaluate belief in human evolution showed that
Blacks and Latinos had large negative coefficients that were statistically significant.
Zigerell (2010) used an OLS regression and found a negative regression coefficient on
African American (-.05).
24
The 2008 GSS question on ethnicity allowed the respondent to indicate one or more races they
considered themselves to be. The response is coded as the first race they mention.
43
geography = f {Middle Atlantic (+); E. North Central (-); W. North Central (+);
South Atlantic (-); E. South Central (-); W. South Central (-); Mountain (-); Pacific (+);
population size (+)}
I included in my model the United States Census divisions in which the surveys
took place (New England division was the excluded variable). My expectation was that
census divisions with mostly Democrat-leaning blue states 25, including the Pacific and
Middle Atlantic division, would show a positive direction of effect on public knowledge
of science. In the remaining census divisions dominated by red states I expected a
negative direction of effect on scientific knowledge due to higher political and religious
conservatism. I expected large, negative effects in the Bible Belt, which is a socially
conservative area of the south-eastern and south-central United States dominated by
evangelical Protestantism. There is some evidence that political and religious
conservatism is negatively associated with public knowledge of science (Nisbet &
Goidel, 2007; Susteric, n.d.). Gibson (2004) showed that a state’s religious environment
is a good predictor of science educational policies. I believe the culture in which a
person lives may act to further enhance and reinforce one’s core values and beliefs which
in turn may impact public knowledge of science.
While I did not find any literature examining the influence of urban versus rural
settings on public knowledge of science, I expected the size of the community in which
25
The terms blue states and red states came into usage in the 2000 general election and denote
states in which residents tend to vote for either Democratic or Republican candidates for the
Presidency, respectively.
44
the GSS interview occurred to show a positive direction of effect on scientific literacy
because of the greater opportunity in an urban environment to access various sources of
scientific information. I included a variable called population size squared in my
quadratic specification of the regression model to see if there was a non-linear
relationship of size of community with public knowledge of science. The units of
population size are expressed in terms of the size of the community to the nearest 1000 in
which the GSS survey took place. I used calculus to find the size of the community in
which the maximum benefit occurs from living in a more populated area in terms of
percentage of correct responses.
Political Factors
political = f {Republican (-); conservative (-)}
I expected political party identification and ideology to be associated with public
knowledge of science. I expected Republican to have a negative direction of effect
compared to Democrats and Independent Party members, which are the excluded
variables. I also expected conservative to show a negative direction of effect compared
to moderate and liberal ideology, which are excluded variables.
I believe Republicans and conservatives may have core values and religious
beliefs that may act as perceptual filters of scientific information. In Chapter 1, I
described how the modern American conservative movement has increasingly dominated
the Republican Party platform and has co-opted the Religious Right to achieve its
political goals (Eckolm, 2011; Gibson, 2004; Gross, 2006; Mooney, 2005). Specter
45
(2009) provides many examples of what he calls denialism of scientific information that
does not comport with one’s deeply held beliefs.
Religious Factors
religion = f {Bible believer (-)}
There are different views of the Bible which range from it being a collection of
fables to it being the inerrant, literal Word of God, absent of errors in every conceivable
way. There is also an intermediate view that the Bible was written by men divinely
inspired but that not everything in the Bible is the literal Word of God. I created a
dummy variable called Bible believer to represent respondents who believed the Bible
was the literal Word of God or the inspired Word of God, since these beliefs are both
strong indications of religious faith. The variable Bible believer shows effects on the
dependent variable relative to respondents who said the Bible is a book of fables (the
excluded variable).
I expected the variable Bible believer to have a negative direction of effect on
public knowledge of science. Zigerell (2010) performed an OLS regression using data
from the 2008 GSS and found a significant, negative coefficient (-.12) on the variable
called literal view of the Bible. Gibson (2004) showed evidence of what he called
“morality policy” on issues like abortion, gay rights, and evolution. He found that a
state’s religious environment is a good predictor of how well evolution is included in
public school’s curriculum. Miller, Scott, and Okamoto (2006) showed the public
acceptance of evolution in the United States was second from the bottom of 34 countries
46
surveyed in 2005; the authors believe it was due to widespread fundamentalism and
political influence on science, a subject extensively discussed by Mooney (2005).
Susteric (n.d.) also found negative relationships between belief in evolution and the
variables fundamentalist Christian and conservative. Nisbet and Goidel (2007) found a
negative relationship between religiosity and support for stem cell research.
Alternative Specification of the Model and Interaction Variables
Regression analysis requires an equation to be linear in the coefficients, according
to the Classical Assumptions (Studenmund, 2006). It is not a violation of the Classical
Assumptions to use alternative functional forms of a model. Therefore, I tested a
quadratic model by including an additional independent variable called population size
squared in my model. I also examined the potential interaction of variables including:
•
Bachelor degree and Bible believer to see if a college education would change
the effect of religiosity on the dependent variable;
•
Bachelor degree and Hispanic to see if a college education would change the
effect of religiosity on the dependent variable;
•
High income and Hispanic see if a relatively high level of income would
change the effect of Hispanic on the dependent variable;
Multicollinearity
Perfect multicollinearity occurs when a linear correlation exists between two or
more of the independent variables of the model violating Classical Assumption VI, which
states, “no explanatory variable is a perfect linear function of any other explanatory
47
variables” (Studenmund, 2006, p. 246). Perfect in this sense means a perfect linear
relationship between variables. Perfect multicollinearity is generally easy to avoid on a
logical basis in the specification of the variables; however, imperfect multicollinearity
can be a little more challenging to detect but can still be corrected for in the model.
According to Studenmund (2006), the consequences of multicollinearity are:
1.
Coefficient estimates will remain unbiased and centered around the true
values in the population.
2.
There will be an increase in the variance and standard errors of the
estimated coefficients.
3.
t-score values will be lower because of the increase in the standard errors.
4.
Coefficient estimates will vary according to the type of model being
specified.
5.
Fit of the equation and coefficients of variables that are not multicollinear
will not substantially change.
There are two ways to detect multicollinearity. The first method is to examine
simple correlation coefficients between the independent variables (Studenmund, 2006). I
performed a bivariate correlation and show the results in Table 3. If a correlation
coefficient is high between two independent variables (generally, if the absolute value is
>.8) then we can expect multicollinearity may be present. This however, is not a
conclusive test.
48
The second way to detect multicollinearity is to examine the variance inflation
factor (VIF) for each independent variable. As a rule of thumb, a VIF (βi) > 5 is a sign
that multicollinearity is severe (Studenmund, 2006). One solution to correct for
multicollinearity is to drop one or more of the multicollinear variables.
Heteroskedasticity
Heteroskedasticity occurs when Classical Assumption V is violated, which
requires “that observations of the error term are drawn from a distribution that has a
constant variance” (Studenmund, 2006, p. 346). Heteroskedasticity typically occurs in
cross-sectional data where it is not uncommon to have a large range between the highest
and lowest observations from the same time period but across different, for example,
geographic areas with varying population size (Studenmund, 2006, p. 349). In my paper,
I am only referring to pure heteroskedasticity described by Studenmund (2006) in which
he describes the consequences as:
1.
Heteroskedasticity does not cause the estimate of coefficients to be biased
and the distribution of the estimates are still close to the actual population
value.
2.
It effects ordinary least squares estimation of coefficients because
heteroskedasticity causes the dependent variable to vary and the regression
model incorrectly attributes this variation to the independent variables.
3.
The estimate of the standard error is biased, which can cause incorrect tstatistics and inaccurate conclusions from hypothesis tests.
49
One of the ways to test for heteroskedasticity is to perform a Park test
(Studenmund, 2006). The Park test is a procedure to test the residual errors in the model
to see how they vary with what is called the proportionality factor, i.e., an independent
variable you believe varies with the variance of the error term (Studenmund, 2006). To
perform the Park test, I saved the residuals of the quadratic model corrected for
multicollinearity. I then took the natural log of the squared residuals and used these
values as the dependent variable in the regression. I then squared the variable called
population size and took its natural log and used it as my independent variable. I then
performed a regression using these variables and examined the model results for
statistical significance.
Regression Coefficients and Ninety Percent Confidence Interval
I ranked the statistically significant regression coefficients (reported at the 0.10
level of significance) in order of importance as measured by their magnitude, i.e., causal
factors that had the greatest influence; second, third, fourth, etc. on the dependent
variable. I also describe the 90% confidence interval for each significant regression
coefficient. The 90% confidence interval tells the range in which the regression
coefficient will occur 90% of the time (i.e., we can be 90% confident the regression
coefficient will fall within this range) and provides more information than a single point
estimate (Studenmund, 2006). I conclude the chapter with a description of the overall fit
of the model and a simulation of the best-case and worst-case scenario in terms of
maximum effects of independent variables on scientific literacy.
50
Conclusion
This chapter described the source of my data and the analytical methods I used to
examine my question of why there is relatively greater knowledge of science in some
adults. I described how I used OLS regression to examine the effect of independent
variables on public knowledge of science. I described a quadratic specification of my
regression model, how I tested for model error, and the potential interaction between
independent variables. Finally, I described how I evaluated regression coefficients in
terms of their rank order of magnitude and the 90% confidence interval.
In the next chapter, I describe the results of my regression analysis and the
significance, direction, and magnitude of regression coefficients for the quadratic
regression model, the final functional form I chose for subsequent analysis. I describe the
results of my tests for interaction between independent variables. I also describe the 90%
confidence interval for significant coefficients. Finally, I describe the overall fit of the
quadratic model as measured by R-squared value.
51
Chapter 4
RESULTS
Introduction
In this chapter, I discuss the descriptive statistics for the dependent and
independent variables used in my model (Table 2). I then describe the results of my test
for significant correlations between independent variables, which can be a sign of
multicollinearity in the model (Table 3). Following this I describe the results of my OLS
regression and the direction and magnitude of regression coefficients, along with
statistical significance reported at the .10, .05, and .01 levels (Table 4). I describe my test
for interaction variables and my test for error in the model. I then describe the regression
coefficients in order of importance as measured by their magnitude and the 90%
confidence interval of each coefficient (Table 5). Finally, I describe the worst- and bestcase scenarios of science literacy (Table 6) and then segue into my last chapter on
conclusions and recommendations.
Descriptive Statistics and Correlation Matrix
Table 2 shows descriptive statistics for the dependent and independent variables.
The dependent variable, which is the percentage of correct responses to the 26 science
and technology questions, has just over 1000 observations (N = 1045). The sample mean
is 58.58 (meaning 58.58% correct responses) with a standard deviation of 22.50. The
maximum percentage of correct responses is 100 and the minimum is zero. The mean
value can be thought of as the average score of all respondents before controlling for any
52
factors causing variation. According to academic grading in the United States, this
average score falls on the border between a D and an F grade. 26
Generational cohorts comprise the following proportions of the sample: 21%
Traditionalist (excluded variable, not shown), 36% Baby Boomer, 29% Generation X,
and 14% Generation Y (Table 2). Males comprise 46% and married respondents
comprise 48% of respondents. Mean family income of respondents was categorized by
income quintiles (see Chapter 3 for description) and breaks down to 22% very low
income (excluded variable, not shown), 23% low income, 16% moderate, 25% high
income, and 14% very high income. Respondents with less than a high school diploma
(excluded variable, not shown) account for 13% of the respondents, followed by 50%
with a high school diploma, 9% with a junior college degree (i.e., Associate Degree),
18% with a bachelor degree, and 10% having a graduate degree. Ethnic groups consisted
of 78% white (excluded variable, not shown), 14% Black, 1% American Indian (includes
native Alaskans), 3% Asian, and 4% Hispanic. Census divisions in which the 2008 GSS
took place indicate the percentage of surveys that took place in each division and were
returned as 4% New England (excluded variable, not shown), 13% Middle Atlantic, 17%
E. North Central, 6% W. North Central, 22% South Atlantic, 5% E. South Central, 10%
W. South Central, 8% Mountain, and 15% Pacific. The variable population size has a
mean value of 368.9, which is the population size of the location in which the 2008 GSS
took place (measured in thousands, the mean population size is 368,900). Republican
26
http://en.wikipedia.org/wiki/Academic_grading_in_the_United_States
53
comprises 26% of the sample; Democrats and Independent Party (excluded variables, not
shown) comprise 74% of respondents. Conservative comprises 20% of the sample and
liberal and moderate ideology (excluded variables, not shown) comprise 80%. Finally,
Bible believer comprises 79% of the sample; the remaining 21% of respondents believe
the Bible is a book of fables (excluded variable, not shown).
Table 3 is a correlation matrix I constructed by performing a bivariate correlation
between independent variables using PASW® Statistics 18. This is a simple, but not
conclusive test for multicollinearity in the model as described in Chapter 3. By
inspection it can be seen there are no correlation coefficients with an absolute value
greater than 0.8 that are statistically significant, based on Pearson’s r test. I later describe
a more definitive test for multicollinearity by examining the Value Inflation Factor or
VIF.
Regression Model Analysis
I first performed an OLS regression analysis on a linear model and found
evidence that age, gender, income, education, ethnicity, and religion were significant
factors (results not shown). I then performed an OLS regression analysis on a quadratic
specification of my regression model, which included the squared value of population
size. The quadratic model showed that population size and population size-squared had
statistically significant regression coefficients. The R-squared value for the quadratic
model was also slightly higher compared to the linear model. 27 Based on these findings,
27
I describe the meaning and results of R-squared later in this chapter.
54
and on my theory that population size has a non-linear effect on the dependent variable, I
settled on the quadratic model for further analysis.
The regression analysis showed evidence of an age effect on scientific literacy
(Table 4). The independent variables Baby Boomer, Generation X, and Generation Y
each were each significant and had positive regression coefficients, relative to the
excluded variable, the Traditionalist generation born between 1900 and 1945.
Furthermore, there was an increase in the magnitude of coefficients from Baby Boomer,
to Generation X, and to Generation Y. This supports my thesis that younger generations
would have relatively higher scientific literacy compared to Traditionalists. Baby
Boomer had a positive regression coefficient of 7.121, which is the change in the
dependent variable (percentage of correct answers) attributable to the independent
variable Baby Boomer, controlling for other factors. 28 This coefficient means Baby
Boomers got 7.121% more questions right (mean of 1.851 questions) compared to
Traditionalists. Generation X had a positive coefficient value of 8.976, meaning this
group got 8.976% more right (mean of 2.333 questions) than Traditionalists. Generation
Y had an even higher coefficient of 12.558, 12.558% more questions right than
Traditionalist (mean of 3.265 questions). This is an interesting pattern of increasing
scientific literacy in younger generations I theorized would be found in the model.
I also found a gender effect in the regression model. The coefficient on the male
dummy was 7.283 and in the positive direction. This result means males answered
28
In equation form, this looks like: percentage of correct responses = 37.220 + 7.121*(Baby
Boomer).
55
7.283% more questions correctly compared to females, the excluded variable (mean of
1.893 questions). This finding supports my thesis that males would have higher scientific
knowledge than females. Zigerell (2010) performed an OLS regression using data from
the 2008 GSS and found the female dummy variable had a regression coefficient of -.08.
This means females, as compared to males, got 8% more wrong answers on the 16-item
science quiz (mean of 2.08 questions). Studies by Miller (1983), the National Science
Foundation (2010), and Susteric (n.d.) also found men scored higher than women on
scientific questionnaires, but there was not a theoretical basis described for these results
nor was Susteric’s (n.d.) result statistically significant. I believe females may still be
subject to cultural stereotypes portraying them as being inferior to men at math and
science, which becomes a self-fulfilling societal expectation. It also may be that more
men are employed in scientific and technical careers than women and have the relevant
college training.
I did not find a significant effect in the model associated with being married, but I
was not too surprised by this result (Table 4). I believe I had a plausible theory why it
might be an important factor so I included it in the model. I theorized married
respondents would perform better on the science questions due in part to having a shared
pool of life experience, interests, and other information that would available to one’s
spouse.
The results show an income effect on scientific literacy. Moderate, high, and very
high income were each significant and had positive coefficients (Table 4). These effects
56
are in comparison to the excluded variable which was very low income, i.e., the first
quintile of income (<$20,453 per year). Furthermore, there was an increase in the
magnitude of coefficients for each income category. Moderate income had a coefficient
of 6.376, meaning 6.376% more questions were answered right compared to the very
low-income group (mean of 1.657 questions). High income had a coefficient of 7.987,
meaning 7.987% more questions were answered correctly (mean of 2.076 questions).
Very high income had a coefficient of 10.551, meaning 10.551% more right answers
compared to the very low-income group (mean of 2.743 questions). These results
support my thesis that higher levels of income, compared to the very low-income
category, would be associated with higher scientific literacy. These results might be
explained by the greater diversity of cultural experiences available to high-income
earners.
The results also show an education effect on scientific literacy. High school
diploma, junior college degree, bachelor degree, and graduate degree were each
significant and had positive coefficients (Table 4). These effects are in comparison to the
excluded variable, respondents having less than a high school diploma. There was also
an increase in the magnitude of coefficients as the level of educational attainment
increased. High school diploma had a positive regression coefficient of 11.283, meaning
high school graduates had 11.283% more right answers (mean of 2.933 questions)
compared to respondents without a high school diploma. Junior college degree had a
positive regression coefficient of 18.605 or 18.605% more right (mean of 4.837
57
questions) compared to respondents without a high school diploma. Bachelor degree had
a positive regression coefficient of 26.210 or 26.210% more right (mean of 6.814
questions) compared to respondents without a high school diploma. Finally, graduate
degree had a positive coefficient of 29.776 or 29.776% more right (mean of 7.741
questions) compared to respondents without a high school diploma. These results
support my thesis that educational attainment would have a positive effect on scientific
literacy.
It is interesting to note the benefit of high school diploma to scientific
understanding as indicated by its regression coefficient (11.283 or 11.283% more correct)
compared to respondents with less than a high school diploma. This result suggests that
policy intervention might include programs aimed at keeping kids in school and
increasing the national high school graduation rate.
I also found an ethnicity effect. Black, American Indian, and Asian each had
negative coefficients that were significant (Table 4). These effects are in comparison to
the excluded variable, White respondents. These comparisons tell us what the effects are
of race, controlling for all other factors. In other words, if two respondents were exactly
the same in all characteristics except one is Black and one is White, for example, the
regression coefficient on Black tells the magnitude and direction of effect that being
Black versus White has on scientific literacy. Black had the highest magnitude
coefficient among ethnic variables. The coefficient on Black was negative and equal to
-16.323, meaning 16.323% more wrong answers compared to White (mean of 4.243
58
questions), followed by Hispanic with a coefficient of -13.078, meaning 13.078% more
wrong compared to White (mean of 3.400 questions). American Indian had a coefficient
-9.472, meaning 9.472% more wrong compared to White (mean of 2.462 questions).
Finally, Asian had a coefficient -6.347, meaning 6.347% more wrong answers compared
to White (mean of 1.650 questions). I speculate these results may be explained in part
due to a lower quality of education received by these ethnic groups in economically
disadvantaged areas. Proficiency in the English language may also be acting as a partial
barrier to understanding the survey questions, particularly for the Asian ethnic group.
Susteric (n.d) constructed a logit model with belief in evolution as the dependent variable
and included Black and Latino ethnic groups as independent variables, along with factors
including religiosity and education, to examine how the various factors affect a person’s
belief in evolution. 29 Susteric’s (n.d.) paper was not clear on how regression coefficients
were estimated and determined to be significant; however, it was reported that Black and
Latino respondents had significant negative coefficients, just as I have found in my
model.
The model results did not show any significant regression coefficients for
geographic variables, based on United States Census divisions (Table 4). The South
Atlantic division variable had a VIF greater than 5 but is of very little concern because
there was no other included variable closely correlated with South Atlantic, so there was
29
In a logit regression the coefficient on the dependent variable is the probability that it is
equal to 1.
59
not any threat of multicollinearity. I therefore did not drop it from the model and
corrections were unnecessary.
Model results also did not show a political effect for either the Republican or the
conservative variable (Table 4). This was a surprising result suggesting party
identification and ideology do not provide good estimates of percentage of correct
responses to the science questions in the 2008 GSS after controlling for other factors.
The model showed a significant religion effect. The coefficient on Bible believer
was -5.331, meaning Bible believers answered 5.331% more questions wrong (mean of
1.386 questions) compared to respondents believing the Bible is just a book of fables, the
excluded variable (Table 4). Zigerell (2010) performed an OLS regression using data
from the 2008 GSS and found a negative coefficient (-.12) on the independent variable
called literal view of the Bible and its effect on scientific knowledge. This coefficient
translates into 12 percentage points less, or about two questions out of 16 possible (mean
of 1.92 questions). This coefficient is in the same negative direction, but larger in
magnitude than the regression coefficient on Bible believer included in my model.
Susteric’s (n.d.) logit model had a regression coefficient on fundamentalist Christian that
was also in the negative direction.
Finally, my model showed a significant population size effect. Population size
and population size squared were statistically significant, however, with very small,
coefficients (0.003 and -5.171E-7, respectively), meaning there was a non-linear effect of
population size on the dependent variable. The combined effect of population size and
60
population size squared was determined by using calculus to obtain the first derivative in
order to find where the slope of the combined effect of population size was equal to zero.
I then solved for population size. The steps in this calculation were as follows:
(1) population effect = 0.003(population size) – 0.0000005(population size)2
(2) d(population effect)/dp = 0.003(population size) - 0.0000005(population size)2
= 0.003 - (2)(0.0000005)(population size)
= 0.003 – 0.00001(population size) => the first derivative;
(3) setting the first derivative equal to zero, I solved for population size:
0.003 – 0.00001(population size) = 0
0.00001(population size) = 0.003
population size = 3000 (which, multiplied by 1000, equals 3 million).
The coefficient of population size squared becomes -1.03E5 after taking the first
derivative. So for every 10,000 increase in population size (recall, units of population
size are expressed in 1000s), the positive effect of population size on the dependent
variable diminishes and the slope becomes zero at a population size of 3 million people.
This suggests that up to a population size of three million, we can expect to see
increasing scientific understanding. When population size exceeds three million people,
the effect of population size on the dependent variable has a diminishing effect. This is
consistent with my theory that living in a more urban setting provides more opportunities
for education and interaction among people and an exchange of ideas, up to a point, and
61
then larger communities start seeing attributes characteristic of densely populated areas
(e.g., higher ethnic diversity, economically disadvantaged areas, etc.).
Interaction Variables
I performed three tests to see if there were interactions between certain
independent variables that in turn had effects on percentage of correct responses. The
tests for interaction between variables and my rationale were as follows:
•
Bachelor degree x Bible believer to see if a college education would change
the effect of religiosity on the dependent variable. I expected the percentage
of correct responses to increase. I also expected the coefficient on Bible
believer to decrease and potentially become insignificant;
•
Bachelor degree x Hispanic to see if a college education would change the
effect of Hispanic on the dependent variable. I expected the percentage of
correct responses to increase. I also expected the coefficient on Hispanic to
become less negative and potentially become insignificant;
•
High income x Hispanic to see if a relatively high level of income would
change the effect of Hispanic on the dependent variable. I expected the
percentage of correct responses to increase. I also expected the coefficient on
Hispanic to become less negative and potentially become insignificant.
•
None of these interaction variables were found to be significant (results not
shown).
62
Multicollinearity and Heteroskedasticity
As previously described in this chapter, the South Atlantic division variable had a
VIF greater than 5 but since there was no other included variable closely correlated with
South Atlantic, there was not any threat of multicollinearity. I therefore did not drop it
from the model and corrections for multicollinearity were unnecessary.
To test for heteroskedasticity, I performed the Park test. I saved the residuals of
the quadratic model and then took the natural log of the squared residuals and used these
values as the dependent variable in a new regression model. I then took the natural log of
population size and used it as my independent variable. I then performed a regression
using these variables and examined the model results for statistical significance. The
coefficient on the natural log of population size was not statistically significant and,
therefore, indicates there is no heteroskedasticity in the quadratic model (results not
shown).
Regression Coefficients and Ninety Percent Confidence Intervals
I ranked the statistically significant regression coefficients (reported at the 0.10
level of significance) in order of importance as measured by their magnitude, i.e., causal
factors that had the greatest influence; second, third, fourth, etc. on the dependent
variable (Table 5). I also describe the upper and lower bound of the 90% confidence
interval (CI) for each regression coefficient. The 90% CI tells the range in which the
regression coefficient will occur 90% of the time (i.e., we can be 90% confident the
regression coefficient will fall within this range) and provides more information than a
63
single-point estimate (Studenmund, 2006). I will discuss the implication of these results
in my conclusions and recommendations chapter where I make suggestions for policy
interventions to improve public knowledge of science.
Graduate degree, bachelor degree, and junior college degree had the top three
highest regression coefficients, which fall within the range of 29.7 to 18.6 (Table 5).
This suggests higher education is an important factor in explaining variations in scientific
literacy. Graduate degree had a positive coefficient of 29.776 or 29.776% more correct
responses compared to respondents without a high school diploma (mean of 7.741
questions). The upper bound of the 90% CI for graduate degree was an astonishing
34.155 which means 34.155% more correct responses (8.880 questions) compared to
respondents lacking a high school diploma. The lower bound of the 90% CI was 25.397
or 25.397% (6.603 questions). I can therefore state with 90% confidence that a
respondent with a graduate degree will answer between 6.603 and 8.880 more questions
correctly compared to respondents lacking a high school diploma. Bachelor degree had a
positive regression coefficient of 26.210 or 26.210% more right answers (mean of 6.814
questions) compared to respondents lacking a high school diploma. The upper bound of
the 90% CI was 29.799 meaning 29.799% more correct (7.747 questions) and the lower
bound was 22.622 meaning 22.622% more correct answers (5.881 questions). Again, I
can therefore state with 90% confidence that a respondent with a bachelor degree will
answer between 5.881 and 7.747 more questions correctly compared to respondents
lacking a high school diploma. Junior college was next in magnitude with a positive
64
regression coefficient of 18.605 or 18.605% more correct answers (mean of 4.837
questions) compared to respondents without a high school diploma. The upper bound of
the 90% CI was 22.589, meaning 22.589% more correct answers (6.731 questions) and
the lower bound was 14.620 or 14.620% more correct answers (3.801 questions).
The next group of variables in rank order of magnitude was Generation Y, high
school diploma, and very high income, with coefficients that fall in the range 12.5 to
10.5. Generation Y had a coefficient of 12.558, meaning 12.558% more correct answers
(mean of 3.265 questions) compared to the excluded category, Traditionalist. The upper
bound of the 90% CI was 15.911, which was 15.911% more correct answers (4.136
questions) while the lower bound was 9.204 or 9.204% more correct answers (2.393
questions). High school diploma follows next in rank order of magnitude with a positive
regression coefficient of 11.283, meaning high school graduates had 11.283% more right
answers (mean of 2.933 questions) compared to respondents without a high school
diploma. The upper bound of the 90% CI was 12.214, meaning 12.214% more correct
answers (3.175 questions) while the lower bound was 8.352 or 8.352% more correct
answers (3.175 questions). The very high-income coefficient was 10.551, meaning
10.551% more right answers (mean of 2.743 questions) compared to the very low-income
group (Table 5). The upper bound of the 90% CI was 14.231, meaning 14.231% more
correct answers (3.700 questions) while the lower bound was 6.872 or 6.872% more
correct answers (1.786 questions). These results support my theory that age, education,
and income would each be significantly associated with scientific literacy.
65
The next group of variables in rank order of magnitude was Generation X, high
income, male, Baby Boomer, and moderate income with coefficients that fall in the range
of 8.9 to 6.3 (Table 5). Generation X had a positive coefficient value of 8.976, meaning
this group got 8.976% more right (mean of 2.333 questions) compared to Traditionalists.
The upper bound of the 90% CI was 11.750, meaning 11.750% more correct answers
(3.055 questions) while the lower bound was 6.202 or 6.202% more correct answers
(1.612 questions). High income is the next highest in magnitude with a coefficient of
7.987, meaning 7.987% more correct answers (mean of 2.074 questions) compared to the
very low-income excluded category. The upper bound of the 90% CI was 11.001,
meaning 11.001% more correct answers (2.860 questions) while the lower bound was
4.973 or 4.973% more correct answers (1.292 questions). Male had a positive coefficient
of 7.283, meaning that compared to the excluded variable female, males answered
7.283% more questions correctly (mean of 2.035 questions). The upper bound of the
90% CI was 9.055 which means 9.055% more correct answers (2.354 questions) while
the lower bound was 5.510 or 5.510% more correct answers (1.326 questions). Baby
Boomer had a positive coefficient of 7.121, meaning 7.121% more correct answers (mean
of 1.851 questions) compared to Traditionalists. The upper bound of the 90% CI was
9.726, meaning 9.726% more correct answers (2.528 questions) while the lower bound
was 4.517 or 4.517% more correct answers (1.174 questions). Finally, moderate income
had a coefficient of 6.376, meaning 6.376% more questions were answered correctly
(mean of 1.657 questions) compared to the very low-income group. The upper bound of
66
the 90% CI was 9.413, meaning 9.413% more correct answers (2.447 questions) while
the lower bound was 3.338 or 3.338% more correct answers (0.878 questions).
The next variables I discuss have negative coefficients, meaning the effect on the
dependent variable is negative. Bible believer, Asian, and American Indian had
coefficients that fall within the range of -5.3 to -9.4 (Table 5). Bible believer had a
negative coefficient of -5.331, meaning 5.331% more questions wrong (mean of 1.386
questions) compared to the excluded variable of Bible is just a book of fables. The upper
bound of the 90% CI was -2.950, meaning 2.950% more wrong answers (0.767
questions) while the lower bound was -7.713, meaning 7.713% more questions wrong
(2.005 questions). Asian had a coefficient -6.347, meaning 6.347% more wrong answers
compared to White, the excluded variable (mean of 1.65 questions). The upper bound of
the 90% CI was -1.149, meaning 1.149% more wrong answers (0.38 questions) while the
lower bound was -11.545, meaning 11.545% more wrong answers (3.001 questions).
American Indian had a negative coefficient of -9.472, meaning 9.472% more wrong
answers compared to White, the excluded variable (mean of 2.462 questions). The upper
bound of the 90% CI was -2.596, meaning 2.596% more wrong answers (0.674
questions) while the lower bound was -16.349, meaning 16.349% more questions wrong
(4.250 questions).
The last two ethnic variables with negative regression coefficients, notable
because of their large magnitudes, are Hispanic and Black. Hispanic had a coefficient of
-13.078, meaning 13.078% more wrong answers compared to White, the excluded
67
variable (mean of 3.400 questions). The upper bound of the 90% CI was -8.298, meaning
8.298% more wrong answers (2.157 questions) while the lower bound was -17.858,
meaning 17.858% more questions answered wrong (4.643 questions). Black had the
highest magnitude negative coefficient equal to -16.323, meaning 16.323% more wrong
answers compared to White, the excluded variable (mean of 4.243 questions). The upper
bound of the 90% CI was -13.687, meaning 13.687% more wrong answers (3.558
questions) while the lower bound was -18.958, meaning 18.958% more questions
answered wrong (4.929 questions).
I speculate the effects of ethnicity described above may be explained in part due
to a lower quality of education received by these ethnic groups compared to Whites or
that perhaps the English language is acting as a partial barrier to understanding the survey
questions, particularly for respondents in which English is spoken as a second language.
The regression coefficient on population size was positive (.003) whereas the
coefficient on population size squared was negative (-5.171E-7), meaning population size
squared is having a diminishing effect on the dependent variable as population size
increases. As described earlier in this chapter, population size reaches a local maximum
(with a slope of zero) at three million people. This is the size of the population in which
this variable reaches its maximum effect on the dependent variable.
Overall Fit of the Model
The R-squared value (also written as R2) is called the coefficient of determination
and must be ≤ 0 R2 ≤ 1. The coefficient of determination measures how well the
68
estimated regression equation fits the sample data and explains variation in the dependent
variable, called the “goodness of fit.” The closer R2 is to one, the better the fit. The Rsquared for my model was 0.497 meaning that almost 50% of the variation in scientific
literacy was explained by the model.
We are cautioned by Studenmund (2006) never to try to maximize the R-squared
value to obtain a better fit for our model. A regression model should be built upon a
logical, theoretical framework using available data and information and common sense
(Studenmund, 2006). It is also important from the very beginning to ask the right
questions, be knowledgeable about the policy context, and carefully review the data.
In summary, I found evidence confirming my theory that socioeconomic factors
and religious factors are associated with public understanding of science. I found age,
gender, income, education, ethnicity, and religion were significant factors explaining
variation in the percentage of correct responses. I found no evidence of significant
effects that could be attributed to being married, to U.S. Census divisions, or due to
political factors as represented by Republican Party and conservative ideology. I also
found no evidence of interaction variables for the combinations of independent variables
I tested.
Simulations of the best-case and worst-case scenario in terms of maximum effects
of independent variables on scientific literacy are shown in Table 6. This table can be
thought of as depicting respondent profiles of the highest and the lowest achievers. The
best-case scenario for highest achievement consisted of a Generation Y, White male,
69
having a very high income and graduate degree, and being a non-Bible believer living in
a community of 3 million people. Conversely, the worst-case scenario for lowest
achievement consisted of a Traditionalist, Black female, having a very low income and
lacking a high school diploma, and being a Bible believer living in a community of 1000
people. These are interesting profiles that gave me ideas for policy interventions I
describe in my final chapter on conclusions and recommendations.
70
Chapter 5
CONCLUSIONS AND RECOMMENDATIONS
The environmental and social problems we face as a nation and as members of the
world community are daunting. Climate change, extreme weather events, scarcity of
fresh water, loss of biodiversity, perpetual war, economic recession, human disease, and
overpopulation are grand problems requiring grand solutions. Conflicts between nations,
political parties, interest groups, economic classes, and ethnic and religious factions will
likely increase as competition for limited natural resources intensify, further impeding
progress toward solutions. There is an urgent need for “believers” on all sides to get out
of their silos, out of their entrenched positions, and stop trying so feverishly to promote
their own particular view as preeminent. Communication and respect needs to be
improved between different groups to foster collaborative problem solving.
To develop strategies to deal with these highly complex and interrelated issues,
there first has to be agreement that a problem exists and that there are different
approaches to solving problems. Without some level of agreement, it is unlikely there
will be public support and political will to implement adaptation strategies and solutions
to these grand problems. Consensus between various groups will be essential to progress.
I described in my introduction how denialism and the politicization of science are
so pervasive in today’s politically charged environment. There are, for example, some
people who do not believe in climate change, biological evolution, or the need for
childhood vaccinations. Organized interest groups and lobbyists work through the policy
71
and political process for the benefit of certain industries to gain access to decision
makers. It is particularly concerning when the goal of these lobbying activities is to
reduce environmental regulations for the benefit of corporate interests.
Illogical thinking, political ideology, and religious dogmatism are impeding
scientific progress and threatening our future. Public understanding of science will
enable citizens to be better informed on policy issues confronting us as a nation. With a
basic understanding of science and what it means to study something scientifically, voters
will be better informed when they register their preferences at the ballot box for political
candidates and decide on various initiatives related to environmental and social issues
they care about.
Scientific literacy is an essential element of a pluralistic society; it is important for
citizens and policymakers to have a basic understanding of science. Policymakers
especially need to understand what the science is telling them in order to make sound
decisions on critical issues, such as adaption measures to deal with climate change or
public health policies on AIDS. Higher public understanding of science is also associated
with greater support for the work of science and scientific institutions, important for
maintaining America’s competiveness with other nations.
Social scientists have been studying public understanding of science for several
decades, particularly in the United States and the United Kingdom. Scientific literacy is,
however, not easy to define, nor do researchers or political leaders agree upon what
72
people “need to know” when it comes to science. We also do not completely know all
the factors associated with a public understanding of science.
Some of the early research on scientific literacy used survey methodologies that
included a battery of science and technology questions, which were then analyzed to find
associations with socioeconomic and demographic factors. Early work was based on the
deficit model that assumed support for science was a function of the public understanding
of science and its processes. If the gap in knowledge could be filled with scientific
factual knowledge, it was believed the public would be more trusting and supportive of
scientific institutions. In contrast, the contextualist model assumed science knowledge is
necessarily related to time, place, culture, values, and particular circumstances. This
model has spawned research in science communications, use of the media, and the role of
scientists in society.
My thesis was influenced by the deficit model, oriented toward the importance of
knowing science facts and methods, to see if I could find significant factors associated
with scientific literacy using a regression-based analysis. My thesis posited that
socioeconomic, political, and religious factors would be good predictors of scientific
literacy, which I defined as the percentage of correct responses to 26 science and
technology questions in the 2008 GSS. What I specifically wanted to find was the
relative magnitudes of these factors in influencing a person’s scientific literacy. To do
this, I described the independent variables I chose as proxies for each of the broad
factors, performed an OLS regression analysis of cross-sectional survey data, and found
73
statistically significant regression coefficients on some of the independent variables. I
then compared the magnitudes of these statistically significant coefficients to see the
relative influence of each. In this final chapter, I discuss the policy implications of each
of my findings (summarized in italics) and my recommendations for policy interventions.
Respondents with a graduate degree, bachelor degree, associate degree, and a
high school diploma performed better on the observed test of scientific literacy
than respondents lacking a high school diploma.
I was not surprised to find that educational factors had the largest magnitude,
positive effects on scientific literacy. Respondents with a graduate degree would be
expected to answer 29.776% more questions correctly than the excluded group
(respondents lacking a high school diploma) and had the largest magnitude coefficient of
all the independent variables in my model (Table 5). A respondent with a high school
diploma would be expected to answer 11.283% more questions correctly in comparison
to the excluded group indicating the tremendous value of completing a high school
education (Table 5).
My results are consistent with findings of other researchers who found education
was positively associated with scientific knowledge. Smith (1996) found educational
attainment had the highest association with scientific knowledge, both at an individual
level and a national level. Nations with higher levels of economic development,
measured by per capita gross national product, had increased scientific knowledge. The
author suggested this finding may be due to stronger science education standards in the
74
classroom in wealthier nations. Pardo and Calvo (2004) found a highly significant
positive relationship between years of education and scientific literacy in EU member
countries. In each nation, they found that individuals with more than 20 years of
education demonstrated a greater number of correct answers on the Eurobarometer
science quiz compared to those individuals with less than 15 years of education. Zigerell
(2010) performed an OLS regression on 2008 GSS data using percentage correct on a 16item science quiz as the dependent variable, and found the regression coefficients on the
independent variables formal education, high school science courses, and college science
courses were in the positive direction and were statistically significant. Susteric (n.d.)
constructed a logit model with belief in evolution as the dependent variable and found the
independent variable college graduate was statistically significant. The variable postgraduate was also statistically significant and had an even larger effect.
Black, American Indian, Asian, and Hispanic ethnic groups were found to have
lower levels of scientific literacy on the observed test. Males had higher levels of
scientific literacy on the observed test compared to females.
Ethnic factors were shown to have the largest magnitude, negative effects on
scientific literacy in comparison to Whites, the excluded group. Black, Hispanic, and
American Indian ethnic groups had the highest magnitude coefficients in the negative
direction of effect (Table 5). These results are consistent with the findings of other
researchers who also found ethnicity associated with scientific knowledge. Susteric (n.d.)
found the independent variables Black and Latino had negative coefficients that were
75
statistically significant. Zigerell (2010) also found that African American had a
coefficient in the negative direction of effect on scientific literacy. Race has always been
an uncomfortable topic in the United States stifling an open discussion of issues that
disproportionately impact certain groups of people. For example, statistics on high
school graduation rates for ethnic minority groups is troubling. Gender was also found to
be significant with males having higher levels of scientific literacy on the observed test.
Males answered 7.283% more questions correctly than females (Table 5). My results are
consistent with findings of other researchers who found gender associated with scientific
knowledge. Smith (1996) found the variable male had a positive association with
scientific knowledge. Zigerell (2010) made the same finding, but instead of a male
dummy, used a female dummy variable, which had a negative coefficient. Susteric (n.d.)
did not find a significant coefficient on the variable male.
A national report from Education Week and the Editorial Projects in Education
(EPE) Research Center found high school graduation rates increased sharply in 2011 and
reached their highest peak since the 1980s. 30 The national graduation rate for public
schools for the class of 2008 was 71.7%. The report also showed that Latinos, African
Americans, and Native Americans had graduation rates less than the national average
(57%, 57%, and 54%, respectively). Asian Americans and Whites did much better with
83% and 78% graduation rates, respectively. Minority males had a graduation rate of less
30
http://www.edweek.org/media/diplomascount2011_pressrelease.pdf
76
than 50%. Females had a higher graduation rate of 75% compared to males (68%) and it
was higher than the national rate.
As described by the National Assessment of Educational Progress (NEAP) 31 data
from 2009, only 1% of 12th graders performed at the Advanced level in NEAP science
while 21% performed at or above the Proficient level and 60% performed at or above the
Basic level. Black, Hispanic and female 12th graders also underperformed compared to
White, male 12th graders. Other interesting findings were that 12th-grade Asian/Pacific
islanders performed better than Whites and the average score of 12th graders was highest
in suburbs and lowest in city school locations.
I believe the ethnic effect found in my results and described in the literature may
be in part due to contrasts between the dominant culture of White Americans and the
minority subcultures. These minority subcultures lack respect from the dominant culture
and are subject to cultural stereotypes perpetuated on television and in other forms of
media. Cultural pressures likely exist within these minority subcultures to conform to
these dominant stereotypes or risk being ostracized by their own groups, an effect called
“stereotype threat,” which can cause stigmatized groups (including ethnic minorities,
women, and individuals from lower socioeconomic backgrounds) to underachieve on
standardized tests and have poor academic performance (Steele, 1997). It may also be
that Whites take more science and technology classes and advanced placement courses
31
National Center for Education Statistics,
http://nces.ed.gov/pubsearch/pubsinfo.asp?pubid=2011451
77
compared to other ethnic groups, which was not accounted for in the GSS data. The GSS
education data used in my regression model only coded educational attainment of
respondents as categories (i.e., high school diploma, associate degree, bachelor degree,
and graduate degree). There was no differentiation of science, technology, mathematics,
and engineering degrees from liberal arts degrees. Drilling down into specific types of
degrees and types of science courses taken by respondents would further improve my
regression model.
My results, along with EPE and NEAP findings, suggest policy intervention is
warranted and should focus on helping high school students stay in school, earn their
diplomas, and improve college-readiness toward a four-year degree. Improvements to
science curricula and educational performance standards may also be necessary. There
also needs to be focused assistance to Black, Hispanic, American Indian, and female high
school students, particularly in city schools. Assistance could take the form of afterschool programs, tutoring programs, and mentoring with practicing scientists. We need
to get students interested in science and technology at an early age while they are in K-12
grades. Strategies also need to be developed and implemented to further reduce the high
school dropout rate.
Programs need to be developed to help K-12 female students increase their
scientific knowledge and understanding; science education should begin in elementary
school with hands-on experience in outdoor and indoor laboratories. We need to do a
better job overall of educating our future scientists, engineers, business leaders, and
78
public officials. Programs such as Girls RISE (Raising Interest in Science and
Engineering) is a good model for engaging and motivating minority girls in grades six
through 12 and developing the next generation of women professionals (Girls RISEnet,
2012). Girls RISE is a National Museum Network funded by the National Science
Foundation. Its objectives are to:
o
Utilize the national network of science centers and museums to raise
awareness and broaden access for girls underrepresented in STEM. 32
o
Develop linkages between organizations with the common purpose of
increasing the pipeline of minority female engineers.
o
Facilitate translation of gender and diversity research into practice through
a unified training program.
o
Provide ongoing services, access to program materials, and tools to
broaden the ability of science centers to provide relevant and engaging
programming for girls. (Girls RISEnet, 2012, para. 3)
The Girls RISE program works to increase the capacity of science centers and
museums across the nation to engage girls, particularly ethnic minorities, in science
education activities. The program facilitates workshops and activities hosted by
practitioners of science and technology, along with museum staff and also administers
32
STEM is an acronym that refers to science, technology, engineering, and mathematics
professions. For information on the STEM initiative (Kuenzi, 2008).
79
small grants and travel subsidies to assist practitioners and students (Girls RISEnet,
2012).
STEM is a national initiative to increase the number of students, teachers, and
practitioners in the fields of science, technology, engineering, and mathematics in the
United States (Kuenzi, 2008). There are many federal initiatives and funding sources
promoting STEM education programs, including programs directed at primary and
secondary schools, girls and young women, and underrepresented groups including ethnic
minorities.
Generation Y, Generation X, and Baby Boomer respondents performed better
than Traditionalist respondents.
I expected to see age factors associated with scientific literacy. Generation Y had
12.558% more correct answers than Traditionalist and had the largest coefficient among
the generational cohorts. It is interesting to note the relative magnitudes of Baby
Boomer, Generation X, and Generation Y, each in comparison to Traditionalist. I
expected younger generations would have increasingly larger positive effects on
scientific literacy. Given this trend, we might expect to see Generation Z, born since
2000 and also called the Internet Generation (Lancaster & Stillman, 2002) perform even
higher than Generation Y on measures of scientific literacy. However, Generation Z
individuals can only participate in the General Social Survey once they turn 18 years of
age.
80
Smith (1996) hypothesized younger adults would have greater scientific and
environmental knowledge because of new emerging science unknown or not widely
covered when older adults received their education. He found knowledge was higher in
younger adults and statistically significant. Zigerell (2010) also found age was associated
with scientific literacy and was statistically significant.
Given the sharp increase in 2011 high school graduation rates, compared to the
prior two decades, as described in the EPE study, I am cautiously optimistic we may see
increased scientific literacy in Generation Z who earn their high school diplomas and
transition on to college. I do not have any specific policy recommendations to address
the age effect seen in my model. There is no particular problem that needs to be solved,
per se. I believe we may start seeing an increase in scientific literacy as Generation Y
and Generation Z become the next cohorts replacing retiring Baby Boomers.
Very high income, high-income, and moderate-income earners performed better
than very low-income earners.
I expected income would be positively associated with scientific literacy.
Respondents with high income and very high income performed better on the observed
test of scientific literacy (7.987% and 10.551%, respectively) compared to the very lowincome group. These results illustrate, at least with respect to scientific literacy, the
disparity between the very high income and the very low-income earners in the United
States. There are clearly advantages afforded to the very wealthy, an issue that has been
81
gaining a lot of media coverage since the start of the Occupy Wall Street movement on
September 17, 2011, bringing international attention to the issue of income inequality.
My results are consistent with findings of other researchers who found income
was positively associated with scientific knowledge. Smith (1996) hypothesized that
those with higher incomes would have more opportunity to consume more upscale and
educational media. He found educational attainment had the highest association with
scientific knowledge at the national level among nations with higher levels of economic
development, measured by per capita gross national product. The author suggested this
finding may be due to stronger science education standards in the classroom in wealthier
nations. Pedro and Calvo (2004) also found that at the national level, the higher the level
of income, the higher percentage of correct answers on a science literacy test. At the
individual level, Smith (1996) did not find income was significant.
My policy recommendation is to help K-12 grade school students from
economically disadvantaged groups stay in school, earn their diplomas, and improve
college-readiness toward a four-year degree. Assistance could also take the form of afterschool programs, tutoring programs, and helping disadvantaged groups obtain a General
Equivalency Diploma. There also needs to be focused assistance to Black, Hispanic,
American Indian and female high school students, particularly in city schools. We need
to level the playing field and reduce the stark divisions existing between economic
classes and between the educated and uneducated in the United States.
82
Bible believers performed worse on the survey questions compared to nonbelievers.
The Bible believer group exhibited 5.331% more wrong answers compared to
non-believers (Table 5). I was somewhat surprised by this result because I thought
religious dogmatism, represented by Bible believer, would have a much larger negative
coefficient. The coefficient on Bible believer was smaller in magnitude than the
coefficients on the ethnicity variables (Table 5).
My results are consistent with findings of other researchers who found dogmatic
religious beliefs were negatively associated with scientific knowledge. Smith (1996)
hypothesized scientific knowledge would be higher among the less religious. He
believed this was because of the conflict between religious belief and some areas of
science and because the more religious would be less interested and engaged in science.
His model showed scientific knowledge was higher among those who do not believe in
God. Individuals with some form of religiosity, measured by church attendance, belief in
God, Protestant and Catholic, showed equivocal results. Zigerell (2010) found a literal
interpretation of the Bible, compared to persons having less dogmatic views, was
negatively associated with scientific knowledge. However, when controlling for
demographic and educational factors, the differences were greatly lessened. Susteric
(n.d.) found the variables fundamentalist and conservative were negatively associated
with belief in evolution.
83
The negative coefficient on Bible believer may be due to respondents’ being
conflicted by their doctrinal belief. Christians who believe the Bible is inerrant and
literally true may have chosen answers on the 2008 GSS survey consistent with their faith
rather than based on scientific evidence. Three true/false questions in the 2008 GSS may
have been particularly troubling to fundamentalist Christians (see Appendix A): (1)
whether the universe began with a huge explosion (which brings into focus the Big Bang
theory versus a divine fiat of the Creator as an explanation for the universe’s origin), (2)
whether the continents have been moving their locations for millions of years and will
continue to do so (which requires acknowledging the Earth is older than about 6,000
years), and (3) whether or not human beings developed from earlier species of animals
(consistent with evolutionary theory).
A federal advisory committee drew criticism from the White House and science
educators when it decided to omit data on understanding of the Big Bang and evolution
from the NSF 2010 Science and Indicators Report (Bhattacharjee, 2010; Guterbock,
2011). The National Science Board, which oversees the NSF, chose to omit this
information from the report for scientific accuracy arguing that these survey questions are
flawed indicators of scientific knowledge because they force respondents to make a
choice between factual scientific knowledge and their own religious beliefs
(Bhattacharjee, 2010). Jon Miller, a leader in the field of public understanding of science
from Michigan State University, criticized the NSB decision stating, “Evolution and the
big bang are not a matter of opinion. If a person says that the earth really is at the center
84
of the universe, even if scientists think it is not, how in the world would you call that
person scientifically literate? Part of being literate is to both understand and accept
scientific constructs” (Bhattacharjee, 2010, p. 2).
In the 2008 GSS, only 32.3% of respondents answered true to the statement that
the universe began with a huge explosion. Only 44.5% of respondents answered true to
the statement that human beings as we know them today, developed from earlier species
of animals. Guterbock (2011) states it is clear some respondents answered questions
based on their doctrinal belief and cautions that the questions on Big Bang and human
evolution involve complex and situational issues, and it would be a mistake to conclude
respondents have a deficit in scientific knowledge based on answers to these questions.
He advises that if the NSF were in the future to keep these questions in the Science and
Engineering Indicators report, then they should be rephrased as “According to
astronomers,” and “According to evolutionary theory.” This would allow respondents a
way to answer these particular science questions without compromising their religious
beliefs.
Because the effect of Bible believer on scientific understanding in my analysis
appears to be small relative to other factors, I do not recommend specific policy
interventions to address this effect. Further study is, however, warranted to evaluate the
associations between the characteristics of respondents and answers to science questions
that may be troubling to certain religious groups. I also recommend science and religious
85
communities improve communications with each other in an attempt to facilitate greater
appreciation and understanding of different ways of knowing the world.
A dichotomy of science and religion did not used to exist, and there is a legitimate
role for both fields of study in the world today. It is also important to recognize religious
beliefs and cultural factors are very difficult to change in people, whereas education can
be changed earlier in life. More emphasis should be placed on providing better science
education to students than trying to convince the faithful they lack scientific rigor in their
beliefs. This would only lead to further conflict between science and religion.
Population size had a non-linear effect on science understanding, with more
populated communities, up to about three million people, having a higher level of
scientific literacy than less populated communities.
Population size had a statistically significant positive effect on scientific literacy,
although the regression coefficient is rather small in magnitude (Table 5). An interesting
finding was that population size had an increasing positive effect up to a community size
of three million people, beyond which benefits diminished.
These results indicate a positive benefit of living in a more populated area. The
benefits of living in more populated communities may include access to more cultural
and educational experiences, including museums, colleges, and lectures, and also greater
social opportunities. I do not have any policy recommendations dealing with population
size. It is, however, an interesting outcome of the regression model and is why I chose a
86
quadratic functional form. I did not discover any findings in the literature similar to my
own regarding the population size effect.
Republican, conservative, and married respondents and geographic regions were
not found to be statistically significant variables on the observed test of scientific
literacy.
I was very surprised not to find statistical significance for the coefficients on
Republican and Conservative. I assumed conservative party identification and ideology
would have a negative direction of effect on scientific literacy. I was less surprised not to
find statistical significance for the coefficients on the married dummy variable and
geographic region variables. I had plausible explanations why these variables might be
important factors associated with scientific literacy but I could not find any statistical
significance to support my theory.
So is it in our collective interest for citizens to have a basic understanding of
science and its methods of investigation? I absolutely believe it is in the public interest.
But it raises the legitimate question of why emphasize science above other disciplines?
Would not public understanding of American history be just as important as or more
important than science for civic duty? Why not public understanding of sociology?
Valid justifications could be constructed for these subject areas as well. I do not think I
have suggested science is preeminent among all subjects of knowledge, and, as a matter
of fact, I believe knowledge of religion and philosophy are just as important as science
and are equally valid ways of experiencing the world around us. I, however, chose public
87
understanding of science for my thesis topic partly because I have a scientific
background, having received a Master of Science degree in Ecology from the University
of California at Davis, in 1988. I also work professionally as a manager of about 40
environmental scientists in the California Department of Fish and Game. My thesis
research has given me greater appreciation of the quantitative methods available
(including OLS regression) to evaluate policy issues and options dealing with a subject
close to my own professional life experience.
88
APPENDICES
89
APPENDIX A
Science and Technology Questions 33
Variable scianheat: WHICH BODILY FEATURES BE BEST SUITED TO A SMALL
ANIMAL LIVING IN A COLD CLIMATE
Please look at Card 1. The two objects shown there have the same mass, but object B
loses heat more quickly than object A. Which combination of bodily features would be
BEST suited to a small animal that lives in a cold climate and needs to minimize heat
loss?
A) Long ears and a long body.
B) Small ears and a short tail (52.4 %).
C) A long nose and a long tail.
D) A short nose and large ears.
E) A long tail and a short nose.
Variable scibigbang: THE UNIVERSE BEGAN WITH A HUGE EXPLOSION
Now, I would like to ask you a few short questions like those you might see on a
television game show. For each statement that I read, please tell me if it is true or false. If
you don’t know or aren’t sure, just tell me so, and we will skip to the next question.
Remember true, false, or don’t know. The universe began with a huge explosion. (Is that
true or false?) true (32.3%)
Variable sciboyorgr: FATHER GENE DECIDES SEX OF BABY
Now, I would like to ask you a few short questions like those you might see on a
television game show. For each statement that I read, please tell me if it is true or false. If
you don’t know or aren’t sure, just tell me so, and we will skip to the next question.
Remember true, false, or don’t know. It is the father’s gene that decides whether the
baby is a boy or a girl. (Is that true or false?) true (61.9%)
Variable scicondrift: THE CONTINENTS HAVE BEEN MOVING
Now, I would like to ask you a few short questions like those you might see on a
television game show. For each statement that I read, please tell me if it is true or false. If
you don’t know or aren’t sure, just tell me so, and we will skip to the next question.
Remember true, false, or don’t know. The continents on which we live have been
moving their locations for millions of years and will continue to move in the future. (Is
that true or false?) true (76.7%)
33
These are the questions from the General Social Survey of 2008 that I used to construct the
dependent variable. Correct answers followed by the percentage of respondents answering
correctly are shown in italics.
90
Variable scidaynight: WHEN DID MOST ERRORS OCCUR
Please look at Card 3. Day-night rhythms dramatically affect our bodies. Probably no
body system is more influenced than the nervous system. The figure on Card 3 illustrates
the number of errors made by shift workers in different portions of the 24-hour cycle.
Based on the data illustrated in the figure, during which of these time periods did the
most errors occur?
A) 2 A.M. to 4 A.M. (77.3%)
B) 8 A.M. to 10 A.M.
C) 12 P.M. to 2 P.M.
D) 2 P.M. to 4 P.M.
E) 8 P.M. to 10 P.M.
Variable sciearthsun: THE EARTH GOES AROUND THE SUN
Now, I would like to ask you a few short questions like those you might see on a
television game show. For each statement that I read, please tell me if it is true or false. If
you don’t know or aren’t sure, just tell me so, and we will skip to the next question.
Remember true, false, or don’t know. Now, does the Earth go around the Sun, or does
the Sun go around the Earth? Earth around the sun (71.7%).
Variable scielectron: ELECTRONS ARE SMALLER THAN ATOMS
Now, I would like to ask you a few short questions like those you might see on a
television game show. For each statement that I read, please tell me if it is true or false. If
you don’t know or aren’t sure, just tell me so, and we will skip to the next question.
Remember true, false, or don’t know. Electrons are smaller than atoms. (Is that true or
false?) true (51.6%)
Variable scierosion: EXAMPLE OF EROSION
Which one of the following is NOT an example of erosion?
A) The wind in the desert blows sand against a rock.
B) A glacier picks up boulders as it moves.
C) A flood washes over a riverbank, and the water carries small soil particles
downstream.
D) An icy winter causes the pavement in a road to crack (56.0%).
Variable scievolved: HUMAN BEINGS DEVELOPED FROM ANIMALS
Now, I would like to ask you a few short questions like those you might see on a
television game show. For each statement that I read, please tell me if it is true or false. If
you don’t know or aren’t sure, just tell me so, and we will skip to the next question.
Remember true, false, or don’t know. Human beings, as we know them today, developed
from earlier species of animals. (Is that true or false?) true (44.5%)
91
Variable scifishexp1: WHAT IS SCIENTIST TRYING TO FIND OUT FROM THIS
EXPERIMENT
Please look at Card 5. What is the scientist trying to find out from this experiment?
A) If the number of fish in the fish bowl affects the behavior of the fish (40.4%).
B) If the temperature of the fish bowl affects the behavior of the fish.
C) If the temperature and the amount of light affect the behavior of the fish.
D) If the number of fish, the temperature, and the amount of light affect the
behavior of the fish.
Variable scifishexp2: WHY DID YOU CHOOSE THAT ANSWER
A) Because I already know what affects the behavior of fish.
B) Because that is what is allowed to change in this experiment (38.1%).
C) Because that is what stays the same in this experiment.
D) Because that is what the scientist decided to include in this experiment.
Variable scigenes: TRAITS ARE TRANSFERRED OVER GENERATION
Traits are transferred from generation to generation through the…
A) sperm only.
B) egg only.
C) sperm and egg (79.2%).
D) testes.
Variable scigills: HOW FISH GET OXYGEN
How do most fish get the oxygen they need to survive?
A) They take in water and break it down into hydrogen and oxygen.
B) Using their gills, they take in oxygen that is dissolved in water (75.7%).
C) They get their oxygen from the food they eat.
D) They come to the surface every few minutes to breathe air into their lungs.
Variable scigoldfish: EXPERIMENT ABOUT GOLDFISH AND TEMPERATURE
Please look at Card 5. What is the scientist trying to find out from this experiment?
A) If the number of fish in the fish bowl affects the behavior of the fish (59.2%).
B) If the temperature of the fish bowl affects the behavior of the fish.
C) If the temperature and the amount of light affect the behavior of the fish.
D) If the number of fish, the temperature, and the amount of light affect the
behavior of the fish.
Variable scihotcore: THE CENTER OF EARTH IS VERY HOT
Now, I would like to ask you a few short questions like those you might see on a
television game show. For each statement that I read, please tell me if it is true or false. If
you don’t know or aren’t sure, just tell me so, and we will skip to the next question.
92
Remember true, false, or don’t know. First, the center of the Earth is very hot. Is that true
or false? true (82.6%)
Variable scih2olife: WHAT PROPERTY OF WATER IS MOST IMPORTANT
What property of water is most important for living organisms?
A) It is odorless.
B) It does not conduct electricity.
C) It is tasteless.
D) It is liquid at most temperatures on Earth (68.1%).
Variable scilasers: LASERS WORK BY FOCUSING SOUND WAVES
Now, I would like to ask you a few short questions like those you might see on a
television game show. For each statement that I read, please tell me if it is true or false. If
you don’t know or aren’t sure, just tell me so, and we will skip to the next question.
Remember true, false, or don’t know. Lasers work by focusing sound waves. (Is that true
or false?) false (48.7%)
Variable scilftplane: KEY FACTOR THAT ENABLES AIRPLANE TO LIFT
Which of the following is a key factor that enables an airplane to lift?
A) Air pressure beneath the wing is greater than that above the wing (53.8%).
B) Pressure within the airplane is greater than that of the outside.
C) Engine power is greater than that of friction.
D) The plane’s wing is lighter than air.
Variable scilitmstxt: WHY LITMUS PAPER DOESN'T CHANGE COLOR IN MIXED
SOLUTION
A solution of hydrochloric acid (HCl) in water will turn blue litmus paper red. A solution
of the base sodium hydroxide (NaOH) in water will turn red litmus paper blue. If the acid
and base solutions are mixed in the right proportion, the resulting solution will cause
neither red nor blue litmus paper to change color. Explain why the litmus paper does not
change color in the mixed solution. Chemical reaction happens that results in products
that don’t react with litmus paper (19.7%).
Variable sciradioact: ALL RADIOACTIVITY IS MAN-MADE
Now, I would like to ask you a few short questions like those you might see on a
television game show. For each statement that I read, please tell me if it is true or false. If
you don’t know or aren’t sure, just tell me so, and we will skip to the next question.
Remember true, false, or don’t know. All radioactivity is man-made. (Is that true or
false?) false (68.6%)
Variable scisalth2o: OCEAN WATER FOR VEGETABLES
93
Please look at Card 4. A gardener has an idea that a plant needs sand in the soil for
healthy growth. In order to test her idea she uses two pots of plants. She sets up one pot
of plants as shown on the top part of the card. Which one of the pictures on the bottom
part of the card shows what she should use for the second pot?
A) Yes, because there is plenty of ocean water.
B) Yes, because ocean water has many natural fertilizers.
C) No, because ocean water is too salty for plants grown on land (85.0%).
D) No, because ocean water is much more polluted than ocean water.
Variable sciseesand: WHICH PICTURE SHOWS WHAT SHE SHOULD USE FOR
THE 2ND POT
Which ONE of the following should she use for the second pot of plants?
A) Sunlight. Sand and water.
B) Dark cupboard. Sand, soil, and water.
C) Dark cupboard. Soil and water.
D) Sunlight. Sand and soil.
E) Sunlight. Soil and water (53.6%).
Variable scistormtxt: DO YOU SEE LIGHTNING BEFORE HEARING THUNDER
Lightning and thunder happen at the same time, but you see the lightning before you hear
the thunder. Explain why this is so. Light travels faster than sound; light reaches eyes
before sound reaches your ears (44.3%).
Variable sciupbreath: WHY SHORT BREATH AT THE TOP OF A MOUNTAIN
For which reason may people experience shortness of breath more quickly at the top of a
mountain than along seashore?
A) A slower pulse rate.
B) A greater gravitational force on the body.
C) A lower percent of oxygen in the blood (68.5%).
D) A faster heartbeat.
E) A slower circulation of blood.
Variable sciviruses: ANTIBIOTICS KILL VIRUSES AS WELL AS BACTERIA
Now, I would like to ask you a few short questions like those you might see on a
television game show. For each statement that I read, please tell me if it is true or false. If
you don’t know or aren’t sure, just tell me so, and we will skip to the next question.
Remember true, false, or don’t know. Antibiotics kill viruses as well as bacteria. (Is that
true or false?) false (53.4%)
Variable sciweighing: WHICH IS THE BEST METHOD TO REPORT WEIGHT OF
LEAF
94
As part of a laboratory experiment, five students measured the weight of the same leaf
four times. They recorded 20 slightly different weights. All of the work was done
carefully and correctly. Their goal was to be as accurate as possible and reduce error in
the experiment to a minimum. Which of the following is the BEST method to report the
weight of the leaf?
A) Ask the teacher to weigh the leaf.
B) Report the first measurement.
C) Average all of the weights that were recorded (66.4%).
D) Average the highest and lowest weights recorded.
E) Discard the lowest five weights.
95
APPENDIX B
Data Tables
Table 1
Dependent and Independent Variables
Variable Name
Dependent variable:
Percentage of correct
responses
Independent variables:
Socioeconomic factors
Age
Traditionalist (excluded
variable)
Baby Boomer
Generation X
Generation Y
Gender
Female (excluded
variable)
Male
Marital status
Not married (excluded
variable)
Married
Income
Very low income
(excluded variable)
Low income
Moderate income
High income
Very high income
Description
GSS variable name
Calculated percentage of correct
responses to 26 science questions
Computed variable
age
1=Baby Boomer dummy
1=Generation X dummy
1=Generation Y dummy
sex
1= Male dummy
marital
1=Married dummy
income06
1= Low income dummy
1= Moderate income dummy
1= High income dummy
1= Very high income dummy
96
Table 1 (continued)
Variable Name
Education
Less than high school
education (excluded
variable)
high school diploma
junior college degree
bachelor degree
graduate degree
Ethnicity
White (excluded
variable)
Black
American Indian
Asian
Hispanic
Description
1= high school diploma dummy
1= junior college degree dummy
1= BA or BS degree dummy
1= graduate degree dummy
racecen1
1= Black dummy
1= American Indian dummy
1= Asian dummy
1= Hispanic dummy
region
Geographic region
New England (excluded
variable)
Middle Atlantic
E. North Central
W. North Central
South Atlantic
E. South Central
W. South Central
Mountain
Pacific
1= Middle Atlantic dummy
1= East North Central dummy
1= West North Central dummy
1=South Atlantic dummy
1= East South Central dummy
1= West South Central dummy
1= Mountain dummy
1= Pacific dummy
Size of community
Population size
Population size squared
continuous variable
continuous variable
Political factors
Political party
Independent, Democrat
Party (excluded
variables)
Republican
GSS variable name
degree
size
partyid
1= Republican Party dummy
97
Table 1 (continued)
Variable Name
Political ideology
Liberal, moderate
ideology (excluded
variables)
Conservative
Religious factors
Religious ideology
Bible is book of fables
(excluded variable)
Bible believer
Description
GSS variable name
polviews
1 = Conservative ideology dummy
Bible
1= Word of God/Inspired Word
dummy
98
Table 2
Descriptive Statistics
Model
N
Percentage of correct responses
1045
.00
100.00
58.5830
22.50493
Baby Boomer
2013
.00
1.00
.3597
.48002
Generation X
2013
.00
1.00
.2891
.45347
Generation Y
2013
.00
1.00
.1431
.35023
Male
2023
.00
1.00
.4592
.49846
Married
2018
.00
1.00
.4817
.49979
Low income
1774
.00
1.00
.2306
.42131
Moderate income
1774
.00
1.00
.1607
.36731
High income
1774
.00
1.00
.2554
.43618
Very high income
1774
.00
1.00
.1404
.34746
high school diploma
2022
.00
1.00
.4960
.50011
junior college degree
2022
.00
1.00
.0856
.27978
bachelor degree
2022
.00
1.00
.1756
.38055
graduate degree
2022
.00
1.00
.0959
.29459
Black
2012
.00
1.00
.1392
.34620
American Indian
2012
.00
1.00
.0134
.11509
Asian
2012
.00
1.00
.0089
.09418
Hispanic
2012
.00
1.00
.0413
.19892
Middle Atlantic
2023
.00
1.00
.1354
.34228
E. North Central
2023
.00
1.00
.1725
.37792
W. North Central
2023
.00
1.00
.0593
.23628
South Atlantic
2023
.00
1.00
.2190
.41366
E. South Central
2023
.00
1.00
.0479
.21371
W. South Central
2023
.00
1.00
.1038
.30508
Mountain
2023
.00
1.00
.0761
.26526
Pacific
2023
.00
1.00
.1458
.35302
Population size
2023
1.00
8008
368.91
1282.252
Population size squared
2023
1.00 64128064.00
1.7795E6
9.73519E6
Republican
2010
.00
1.00
.2561
.43658
Conservative
1933
.00
1.00
.2043
.40333
Minimum
Maximum
Mean
Std. Deviation
99
Table 2 (continued)
Model
Bible believer
N
1987
Minimum
.00
Maximum
1.00
Mean
.7866
Std. Deviation
.40980
Note that these N values are for the original 2008 GSS variables I used to create dummies for my model.
For example, the GSS variable “age” (N=2013) was used to create the generation dummies shown in the
table. Population size (in thousands) refers to the actual size of the place in which the interview occurred.
It was measured to the nearest 1,000 of the smallest civil division listed by the U. S. Census. For example,
1 = 1000; 10 = 10,000 and so on.
Table 3
Correlation Matrix
Baby
Boomer
dummy
Baby Boomer Pearson
Correlation
Sig. (2-tailed)
N
Generation X
Male
Married
.000
.000
.077
Married
.122**
Moderate
income
.022
High
income
.071**
Very
high
income
.110**
high
school
diploma
-.028
junior
college
degree
.038
bachelor
degree
.002
graduate
degree
.058**
Black
.007
American
Indian
-.033
.000
.000
.356
.003
.000
.208
.088
.932
.009
.771
.136
2013
2013
2013
2010
1767
1767
1767
1767
2012
2012
2012
2012
2002
2002
1
-.261**
.012
.054*
.010
.023
.029
.011
-.050*
.000
.079**
.038
.011
.002
.000
.603
.015
.670
.324
.217
.648
.025
.984
.000
.090
.622
.939
2013
2013
2013
2013
2010
1767
1767
1767
1767
2012
2012
2012
2012
2002
2002
-.306**
-.261**
1
-.021
-.259**
.034
-.023
-.066**
-.092**
.090**
-.021
-.053*
-.119**
.077**
.064**
.000
.000
.339
.000
.151
.330
.006
.000
.000
.336
.017
.000
.001
.004
N
2013
2013
2013
2013
2010
1767
1767
1767
1767
2012
2012
2012
2012
2002
2002
Pearson
Correlation
Sig. (2-tailed)
.039
.012
-.021
1
.046*
-.010
.001
.048*
.049*
-.014
.002
.026
-.004
-.036
.005
.077
.603
.339
.040
.666
.965
.043
.037
.543
.934
.246
.864
.104
.816
N
2013
2013
2013
2023
2018
1774
1774
1774
1774
2022
2022
2022
2022
2012
2012
.122**
.054*
-.259**
.046*
1
-.115**
.004
.238**
.216**
-.049*
.027
.089**
.062**
-.143**
-.057*
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
N
Moderate
income
Male
.039
Low
income
-.109**
2013
N
Low income
Generation Y
-.306**
-.478**
Pearson
Correlation
Sig. (2-tailed)
N
Generation Y
1
Generation X
-.478**
.000
.000
.015
.000
.040
.000
.867
.000
.000
.027
.225
.000
.005
.000
.010
2010
2010
2010
2018
2018
1773
1773
1773
1773
2017
2017
2017
2017
2007
2007
-.109**
.010
.034
-.010
-.115**
1
-.239**
-.321**
-.221**
.118**
-.003
-.105**
-.126**
.068**
.041
.000
.670
.151
.666
.000
.000
.000
.000
.000
.883
.000
.000
.004
.084
1767
1767
1767
1774
1773
1774
1774
1774
1774
1774
1774
1774
1774
1765
1765
.004
**
1
**
**
*
.028
-.008
.017
-.018
-.017
Pearson
Correlation
Sig. (2-tailed)
.022
.023
-.023
.001
-.239
.356
.324
.330
.965
.867
.000
N
1767
1767
1767
1774
1773
1774
1774
-.256
-.177
.047
.000
.000
.048
.243
.745
.485
.452
.474
1774
1774
1774
1774
1774
1774
1765
1765
100
Table 3 (continued)
High income
Pearson
Correlation
Sig. (2-tailed)
N
Very high
income
high school
diploma
Pearson
Correlation
Sig. (2-tailed)
graduate
degree
Black
American
Indian
Generation X
.029
Generation Y
-.066**
Male
.048*
Married
.238**
Low
income
-.321**
Moderate
income
-.256**
High
income
1
Very
high
income
-.237**
high
school
diploma
-.053*
junior
college
degree
.029
bachelor
degree
.132**
graduate
degree
.086**
Black
-.075**
American
Indian
-.062**
.003
.217
.006
.043
.000
.000
.000
.000
.024
.216
.000
.000
.002
.009
1767
1767
1767
1774
1773
1774
1774
1774
1774
1774
1774
1774
1774
1765
1765
.110**
.011
-.092**
.049*
.216**
-.221**
-.177**
-.237**
1
-.139**
.001
.138**
.206**
-.117**
-.037
.000
.648
.000
.037
.000
.000
.000
.000
.000
.953
.000
.000
.000
.118
N
1767
1767
1767
1774
1773
1774
1774
1774
1774
1774
1774
1774
1774
1765
1765
Pearson
Correlation
Sig. (2-tailed)
-.028
-.050*
.090**
-.014
-.049*
.118**
.047*
-.053*
-.139**
1
-.303**
-.458**
-.323**
.046*
.040
.208
.025
.000
.543
.027
.000
.048
.024
.000
.000
.000
.000
.040
.076
N
2012
2012
2012
2022
2017
1774
1774
1774
1774
2022
2022
2022
2022
2011
2011
.038
.000
-.021
.002
.027
-.003
.028
.029
.001
-.303**
1
-.141**
-.100**
.000
-.020
junior college Pearson
degree
Correlation
Sig. (2-tailed)
bachelor
degree
Baby
Boomer
dummy
.071**
.088
.984
.336
.934
.225
.883
.243
.216
.953
.000
.000
.000
.984
.361
N
2012
2012
2012
2022
2017
1774
1774
1774
1774
2022
2022
2022
2022
2011
2011
Pearson
Correlation
Sig. (2-tailed)
.002
.079**
-.053*
.026
.089**
-.105**
-.008
.132**
.138**
-.458**
-.141**
1
-.150**
-.092**
-.031
.932
.000
.017
.246
.000
.000
.745
.000
.000
.000
.000
.000
.000
.162
N
2012
2012
2012
2022
2017
1774
1774
1774
1774
2022
2022
2022
2022
2011
2011
Pearson
Correlation
Sig. (2-tailed)
.058**
.038
-.119**
-.004
.062**
-.126**
.017
.086**
.206**
-.323**
-.100**
-.150**
1
-.062**
-.038
.009
.090
.000
.864
.005
.000
.485
.000
.000
.000
.000
.000
.005
.089
N
Pearson
Correlation
Sig. (2-tailed)
2012
.007
2012
.011
2012
.077**
2022
-.036
2017
-.143**
1774
.068**
1774
-.018
1774
-.075**
1774
-.117**
2022
.046*
2022
.000
2022
-.092**
2022
-.062**
2011
1
2011
-.047*
.771
.622
.001
.104
.000
.004
.452
.002
.000
.040
.984
.000
.005
N
2002
2002
2002
2012
2007
1765
1765
1765
1765
2011
2011
2011
2011
2012
2012
Pearson
Correlation
Sig. (2-tailed)
-.033
.002
.064**
.005
-.057*
.041
-.017
-.062**
-.037
.040
-.020
-.031
-.038
-.047*
1
.136
.939
.004
.816
.010
.084
.474
.009
.118
.076
.361
.162
.089
.035
N
2002
2002
2002
2012
2007
1765
1765
1765
1765
2011
2011
2011
2011
2012
.035
2012
101
Table 3 (continued)
Asian
Pearson
Correlation
Sig. (2-tailed)
Middle
Atlantic
E. North
Central
W. North
Central
South
Atlantic
E. South
Central
W. South
Central
Generation X
.133**
Generation Y
-.014
Male
.010
Married
.039
Low
income
-.049*
Moderate
income
-.021
High
income
.002
Very
high
income
.108**
high
school
diploma
-.097**
junior
college
degree
-.019
bachelor
degree
.112**
graduate
degree
.095**
Black
-.077**
American
Indian
-.022
.000
.000
.530
.653
.079
.041
.385
.936
.000
.000
.390
.000
.000
.001
.320
1995
1995
1995
2005
2000
1759
1759
1759
1759
2004
2004
2004
2004
2005
2005
-.076**
.085**
.082**
.030
-.060**
.045
-.019
-.034
-.066**
-.021
-.001
-.063**
-.067**
-.083**
-.024
.001
.000
.000
.186
.007
.058
.417
.159
.005
.343
.955
.005
.002
.000
.278
N
2002
2002
2002
2012
2007
1765
1765
1765
1765
2011
2011
2011
2011
2012
2012
Pearson
Correlation
Sig. (2-tailed)
-.004
.000
-.013
-.069**
-.023
-.035
-.023
.034
.068**
.012
.003
.011
.053*
.046*
-.008
.872
.992
.570
.002
.292
.142
.343
.158
.004
.596
.897
.621
.018
.038
.708
N
Hispanic
Baby
Boomer
dummy
-.085**
Pearson
Correlation
Sig. (2-tailed)
N
2013
2013
2013
2023
2018
1774
1774
1774
1774
2022
2022
2022
2022
2012
2012
Pearson
Correlation
Sig. (2-tailed)
-.011
-.042
-.025
-.011
-.008
.057*
-.014
-.036
-.019
.009
.006
-.031
-.020
-.016
.004
.609
.057
.253
.615
.715
.015
.556
.131
.419
.691
.796
.159
.380
.465
.860
N
Pearson
Correlation
Sig. (2-tailed)
2013
.043
2013
-.054*
2013
.005
2023
.016
2018
.060**
1774
.034
1774
.006
1774
-.003
1774
-.005
2022
.002
2022
-.017
2022
.033
2022
.003
2012
-.071**
2012
-.029
.054
.015
.823
.462
.007
.152
.789
.893
.840
.929
.446
.142
.876
.001
.188
N
2013
2013
2013
2023
2018
1774
1774
1774
1774
2022
2022
2022
2022
2012
2012
Pearson
Correlation
Sig. (2-tailed)
-.011
.037
-.014
.013
.030
-.022
.029
-.028
-.008
.039
-.025
-.028
.006
.148**
-.010
.633
.101
.543
.548
.174
.346
.221
.247
.743
.081
.257
.215
.785
.000
.667
N
2013
2013
2013
2023
2018
1774
1774
1774
1774
2022
2022
2022
2022
2012
2012
Pearson
Correlation
Sig. (2-tailed)
.017
-.014
-.032
.021
.059**
-.029
-.005
.062**
-.022
.032
.006
-.031
-.010
.018
-.026
.449
.525
.157
.352
.008
.225
.843
.009
.349
.152
.794
.169
.645
.425
.242
N
2013
2013
2013
2023
2018
1774
1774
1774
1774
2022
2022
2022
2022
2012
2012
Pearson
Correlation
Sig. (2-tailed)
-.002
.033
.000
.005
.009
.025
.030
-.033
-.057*
-.030
-.023
-.016
-.034
.051*
.031
.936
.135
.993
.819
.678
.298
.200
.160
.016
.181
.301
.459
.128
.021
.163
N
2013
2013
2013
2023
2018
1774
1774
1774
1774
2022
2022
2022
2022
2012
2012
102
Table 3 (continued)
Mountain
Pacific
Population
size
Republican
Conservative
Biblebeliever
Pearson
Correlation
Sig. (2-tailed)
Baby
Boomer
dummy
-.029
Generation X
.018
Generation Y
.011
Male
-.018
Married
-.016
Low
income
.032
Moderate
income
.011
High
income
.005
Very
high
income
-.032
high
school
diploma
.021
junior
college
degree
.019
bachelor
degree
.034
graduate
degree
-.056*
Black
-.099**
American
Indian
.015
.197
.408
.638
.427
.484
.176
.630
.836
.181
.347
.398
.125
.012
.000
.489
N
2013
2013
2013
2023
2018
1774
1774
1774
1774
2022
2022
2022
2022
2012
2012
Pearson
Correlation
Sig. (2-tailed)
-.017
.012
.086**
.063**
-.076**
-.029
-.035
.013
.029
-.037
.019
.008
-.001
-.101**
.001
.455
.589
.000
.004
.001
.220
.146
.588
.223
.093
.397
.715
.948
.000
.970
N
2013
2013
2013
2023
2018
1774
1774
1774
1774
2022
2022
2022
2022
2012
2012
Pearson
Correlation
Sig. (2-tailed)
-.011
.045*
-.022
-.039
-.031
.015
.001
-.014
.015
-.022
-.029
.025
.029
.132**
.029
.632
.041
.328
.079
.163
.521
.963
.567
.534
.321
.197
.262
.191
.000
.200
N
2013
2013
2013
2023
2018
1774
1774
1774
1774
2022
2022
2022
2022
2012
2012
Pearson
Correlation
Sig. (2-tailed)
-.004
-.018
-.062**
.006
.131**
-.101**
.008
.088**
.160**
-.012
.012
.097**
.004
-.210**
-.049*
.862
.414
.006
.792
.000
.000
.730
.000
.000
.582
.608
.000
.871
.000
.029
N
1965
1965
1965
1972
1968
1734
1734
1734
1734
1971
1971
1971
1971
1962
1962
Pearson
Correlation
Sig. (2-tailed)
.004
-.051*
-.057*
.015
.114**
-.059*
.018
.033
.047
-.009
-.027
.024
-.031
-.076**
-.010
.874
.026
.012
.517
.000
.015
.458
.177
.052
.693
.237
.300
.172
.001
.646
N
1923
1923
1923
1933
1929
1714
1714
1714
1714
1932
1932
1932
1932
1924
1924
-.007
**
**
.008
**
**
.024
**
**
**
-.056*
Pearson
Correlation
Sig. (2-tailed)
-.008
-.036
-.136
.077
.008
.039
-.091
.096
-.096
-.123
.099
.711
.105
.747
.000
.001
.749
.101
.732
.000
.000
.283
.000
.000
.000
.013
N
1978
1978
1978
1987
1984
1754
1754
1754
1754
1986
1986
1986
1986
1977
1977
103
Table 3 (continued)
Baby
Boomer
Generation X
Generation Y
Male
Married
Low income
Moderate
income
Pearson
Correlation
Sig. (2tailed)
N
Middle
Atlantic
-.004
E. North
Central
-.011
W.
North
Central
.043
.001
.872
.609
.054
.633
Asian
Hispanic
-.085**
-.076**
.000
W.
South
Central
-.002
Mountain
-.029
Pacific
-.017
Population
size
-.011
Republican
-.004
Conservative
.004
Biblebeliever
-.008
.449
.936
.197
.455
.632
.862
.874
.711
South
E. South
Atlantic
Central
-.011
.017
1995
2002
2013
2013
2013
2013
2013
2013
2013
2013
2013
1965
1923
1978
Pearson
Correlation
Sig. (2tailed)
N
.133**
.085**
.000
-.042
-.054*
.037
-.014
.033
.018
.012
.045*
-.018
-.051*
-.036
.000
.000
.992
.057
.015
.101
.525
.135
.408
.589
.041
.414
.026
.105
1995
2002
2013
2013
2013
2013
2013
2013
2013
2013
2013
1965
1923
1978
Pearson
Correlation
Sig. (2tailed)
N
-.014
.082**
-.013
-.025
.005
-.014
-.032
.000
.011
.086**
-.022
-.062**
-.057*
-.007
.530
.000
.570
.253
.823
.543
.157
.993
.638
.000
.328
.006
.012
.747
1995
2002
2013
2013
2013
2013
2013
2013
2013
2013
2013
1965
1923
1978
.010
.030
**
-.011
.016
.013
.021
.005
-.018
**
-.039
.006
.015
-.136**
.653
.186
.002
.615
.462
.548
.352
.819
.427
.004
.079
.792
.517
.000
Pearson
Correlation
Sig. (2tailed)
N
Pearson
Correlation
Sig. (2tailed)
N
Pearson
Correlation
Sig. (2tailed)
N
Pearson
Correlation
Sig. (2tailed)
N
-.069
.063
2005
2012
2023
2023
2023
2023
2023
2023
2023
2023
2023
1972
1933
1987
.039
-.060**
-.023
-.008
.060**
.030
.059**
.009
-.016
-.076**
-.031
.131**
.114**
.077**
.079
.007
.292
.715
.007
.174
.008
.678
.484
.001
.163
.000
.000
.001
2000
2007
2018
2018
2018
2018
2018
2018
2018
2018
2018
1968
1929
1984
*
.034
-.022
-.029
.025
.032
-.029
.015
**
*
.008
.015
.152
.346
.225
.298
.176
.220
.521
.000
.015
.749
*
.045
-.035
.041
.058
.142
-.049
.057
-.101
-.059
1759
1765
1774
1774
1774
1774
1774
1774
1774
1774
1774
1734
1714
1754
-.021
-.019
-.023
-.014
.006
.029
-.005
.030
.011
-.035
.001
.008
.018
.039
.385
.417
.343
.556
.789
.221
.843
.200
.630
.146
.963
.730
.458
.101
1759
1765
1774
1774
1774
1774
1774
1774
1774
1774
1774
1734
1714
1754
104
Table 3 (continued)
High income
Very high
income
high school
diploma
junior college
degree
bachelor
degree
graduate
degree
Black
Pearson
Correlation
Sig. (2tailed)
N
Asian
Hispanic
.002
-.034
Middle
Atlantic
.034
E. North
Central
-.036
W.
North
Central
-.003
South
E. South
Atlantic
Central
-.028
.062**
W.
South
Central
-.033
Mountain
.005
Pacific
.013
Population
size
-.014
Republican
.088**
Conservative
.033
Biblebeliever
.008
.936
.159
.158
.131
.893
.247
.009
.160
.836
.588
.567
.000
.177
.732
1759
1765
1774
1774
1774
1774
1774
1774
1774
1774
1774
1734
1714
1754
Pearson
Correlation
Sig. (2tailed)
N
.108**
-.066**
.068**
-.019
-.005
-.008
-.022
-.057*
-.032
.029
.015
.160**
.047
-.091**
.000
.005
.004
.419
.840
.743
.349
.016
.181
.223
.534
.000
.052
.000
1759
1765
1774
1774
1774
1774
1774
1774
1774
1774
1774
1734
1714
1754
Pearson
Correlation
Sig. (2tailed)
N
**
-.021
.012
.009
.002
.039
.032
-.030
.021
-.037
-.022
-.012
-.009
.096**
.000
.343
.596
.691
.929
.081
.152
.181
.347
.093
.321
.582
.693
.000
2004
2011
2022
2022
2022
2022
2022
2022
2022
2022
2022
1971
1932
1986
Pearson
Correlation
Sig. (2tailed)
N
-.019
-.001
.003
.006
-.017
-.025
.006
-.023
.019
.019
-.029
.012
-.027
.024
.390
.955
.897
.796
.446
.257
.794
.301
.398
.397
.197
.608
.237
.283
2004
2011
2022
2022
2022
2022
2022
2022
2022
2022
2022
1971
1932
1986
Pearson
Correlation
Sig. (2tailed)
N
**
.112
**
.011
-.031
.033
-.028
-.031
-.016
.034
.008
.025
**
.024
-.096**
.000
.005
.621
.159
.142
.215
.169
.459
.125
.715
.262
.000
.300
.000
2004
2011
2022
2022
2022
2022
2022
2022
2022
2022
2022
1971
1932
1986
Pearson
Correlation
Sig. (2tailed)
N
.095**
-.067**
.053*
-.020
.003
.006
-.010
-.034
-.056*
-.001
.029
.004
-.031
-.123**
.000
.002
.018
.380
.876
.785
.645
.128
.012
.948
.191
.871
.172
.000
2004
2011
2022
2022
2022
2022
2022
2022
2022
2022
2022
1971
1932
1986
Pearson
Correlation
Sig. (2tailed)
N
-.077**
-.083**
.046*
-.016
-.071**
.148**
.018
.051*
-.099**
-.101**
.132**
-.210**
-.076**
.099**
.001
.000
.038
.465
.001
.000
.425
.021
.000
.000
.000
.000
.001
.000
2005
2012
2012
2012
2012
2012
2012
2012
2012
2012
2012
1962
1924
1977
-.097
-.063
.097
105
Table 3 (continued)
American
Indian
Asian
Hispanic
Middle
Atlantic
E. North
Central
W. North
Central
South
Atlantic
E. South
Central
Pearson
Correlation
Sig. (2tailed)
N
Pearson
Correlation
Sig. (2tailed)
N
Pearson
Correlation
Sig. (2tailed)
N
Pearson
Correlation
Sig. (2tailed)
N
Middle
Atlantic
-.008
E. North
Central
.004
W.
North
Central
-.029
.278
.708
.860
.188
2005
2012
2012
2012
2012
2012
2012
2012
2012
2012
2012
1962
1924
1977
1
-.040
.004
-.029
-.025
-.028
-.043
-.038
-.044*
.161**
.072**
-.011
-.016
-.020
.077
.848
.189
.262
.203
.056
.087
.047
.000
.001
.623
.480
.365
2005
2005
2005
2005
2005
2005
2005
2005
2005
2005
2005
1955
1917
1970
-.040
1
.005
-.055*
-.052*
-.019
-.046*
.069**
-.003
.099**
.058**
-.082**
-.038
-.016
.809
.014
.019
.387
.037
.002
.895
.000
.009
.000
.099
.485
2012
1
2012
-.181**
2012
-.099**
2012
-.210**
2012
-.089**
2012
-.135**
2012
-.114**
2012
-.164**
2012
.340**
1962
-.019
1924
-.060**
1977
-.039
.000
.000
.000
.000
.000
.000
.000
.000
.397
.008
.079
2023
2023
2023
2023
2023
2023
2023
2023
1972
1933
1987
1
**
**
**
**
**
**
*
.015
.011
.013
.048
.501
.619
.560
Asian
Hispanic
-.022
-.024
.320
.077
South
E. South
Atlantic
Central
-.010
-.026
.667
.242
W.
South
Central
.031
Mountain
.015
Pacific
.001
Population
size
.029
Republican
-.049*
Conservative
-.010
Biblebeliever
-.056*
.163
.489
.970
.200
.029
.646
.013
2005
.004
2012
.005
.848
.809
2005
2012
2023
Pearson
Correlation
Sig. (2tailed)
N
-.029
*
**
2005
2012
2023
2023
2023
2023
2023
2023
2023
2023
2023
1972
1933
1987
Pearson
Correlation
Sig. (2tailed)
N
Pearson
Correlation
Sig. (2tailed)
N
-.025
-.052*
-.099**
-.115**
1
-.133**
-.056*
-.085**
-.072**
-.104**
-.061**
-.012
-.004
-.055*
.262
.019
.000
.000
.000
.011
.000
.001
.000
.006
.590
.869
.014
2005
-.028
2012
-.019
2023
-.210**
2023
-.242**
2023
-.133**
2023
1
2023
-.119**
2023
-.180**
2023
-.152**
2023
-.219**
2023
-.123**
1972
.026
1933
.035
1987
.059**
.203
.387
.000
.000
.000
.000
.000
.000
.000
.000
.256
.127
.009
2005
2012
2023
2023
2023
2023
2023
2023
2023
2023
2023
1972
1933
1987
Pearson
Correlation
Sig. (2tailed)
N
-.043
-.046*
-.089**
-.102**
-.056*
-.119**
1
-.076**
-.064**
-.093**
-.049*
.029
.034
.084**
.056
.037
.000
.000
.011
.000
.001
.004
.000
.027
.194
.133
.000
2005
2012
2023
2023
2023
2023
2023
2023
2023
2023
1972
1933
1987
.189
-.055
.014
-.181
.000
-.115
.000
-.242
.000
-.102
.000
2023
-.155
.000
-.131
.000
-.189
.000
-.044
106
Table 3 (continued)
W. South
Central
Mountain
Pacific
Population
size
Republican
Conservative
Bible
believer
Pearson
Correlation
Sig. (2tailed)
N
Asian
Hispanic
-.038
.069**
Middle
Atlantic
-.135**
E. North
Central
-.155**
W.
North
Central
-.085**
South
E. South
Atlantic
Central
-.180**
-.076**
.087
.002
.000
.000
.000
.000
.001
2005
W.
South
Central
1
2012
2023
2023
2023
2023
2023
2023
*
-.003
**
**
**
**
**
**
.047
.895
.000
.000
.422
.527
.256
.000
2023
2023
2023
1972
1933
1987
1
**
-.043
.010
.033
-.044*
.000
.051
.651
.146
.049
2012
2023
2023
2023
2023
2023
2023
2023
2023
2023
1972
1933
1987
.099**
-.164**
-.189**
-.104**
-.219**
-.093**
-.141**
-.119**
1
.007
-.021
-.026
-.094**
.000
.000
.000
.000
.000
.000
.000
.000
.000
.759
.341
.261
.000
2005
2012
2023
2023
2023
2023
2023
2023
2023
2023
1972
1933
1987
Pearson
Correlation
Sig. (2tailed)
N
**
**
**
*
**
**
*
-.018
-.043
.007
1
**
-.027
-.001
.422
.051
.759
.003
.241
.978
.001
.058
.009
.340
-.044
.000
.048
-.061
-.123
.004
-.049
.006
.000
.027
-.098
Biblebeliever
.113**
2005
.000
-.064
Conservative
.026
.161**
.001
-.152
Republican
-.014
Pearson
Correlation
Sig. (2tailed)
N
.000
-.072
Population
size
-.018
-.044
.000
-.131
Pacific
-.141**
Pearson
Correlation
Sig. (2tailed)
N
.072
-.114
Mountain
-.098**
.000
-.119
2023
-.068
2005
2012
2023
2023
2023
2023
2023
2023
2023
2023
2023
1972
1933
1987
Pearson
Correlation
Sig. (2tailed)
N
-.011
-.082**
-.019
.015
-.012
.026
.029
-.014
.010
-.021
-.068**
1
.368**
.135**
.623
.000
.397
.501
.590
.256
.194
.527
.651
.341
.003
.000
.000
1955
1962
1972
1972
1972
1972
1972
1972
1972
1972
1972
1972
1890
1940
Pearson
Correlation
Sig. (2tailed)
N
-.016
-.038
-.060**
.011
-.004
.035
.034
.026
.033
-.026
-.027
.368**
1
.128**
.480
.099
.008
.619
.869
.127
.133
.256
.146
.261
.241
.000
1917
1924
1933
1933
1933
1933
1933
1933
1933
1933
1933
1890
1933
1903
*
**
**
**
*
**
-.001
**
**
1
Pearson
Correlation
Sig. (2tailed)
N
-.055
.059
.084
.113
-.044
-.094
.135
.000
-.020
-.016
-.039
.013
.128
.365
.485
.079
.560
.014
.009
.000
.000
.049
.000
.978
.000
.000
1970
1977
1987
1987
1987
1987
1987
1987
1987
1987
1987
1940
1903
1987
**. Correlation is significant at the 0.01 level (2-tailed).
*. Correlation is significant at the 0.05 level (2-tailed).
107
108
Table 4
Quadratic Model Regression Coefficients
B
Model
(Constant)
(Std. Error)
t
Sig.
VIF Statistics
37.220
10.761
.000***
4.503
.000***
2.173
5.329
.000***
2.132
6.166
.000***
1.745
6.766
.000***
1.044
-.546
.585
1.354
1.312
.190
1.770
3.456
.001***
1.762
4.363
.000***
2.267
4.722
.000***
2.010
6.339
.000***
2.847
7.689
.000***
1.734
12.026
.000***
2.484
11.196
.000***
1.926
-10.197
.000***
1.284
(3.459)
Baby Boomer
7.121
(1.582)
Generation X
8.976
(1.685)
Generation Y
12.558
(2.037)
Male
7.283
(1.076)
Married
-.669
(1.226)
Low income
2.152
(1.640)
Moderate income
6.376
(1.845)
High income
7.987
(1.830))
Very high income
10.551
(2.235)
high school diploma
11.283
(1.780)
junior college degree
18.605
(2.420)
bachelor degree
26.210
(2.179)
graduate degree
29.776
(2.659)
Black
-16.323
(1.601)
109
Table 4 (continued)
B
Model
American Indian
(Std. Error)
t
Sig.
VIF Statistics
-9.472
-2.268
.024**
1.060
-2.011
.045**
1.113
-4.506
.000***
1.167
.512
.609
3.071
.769
.442
3.879
.177
.860
2.342
.448
.654
5.294
-.247
.805
2.194
-1.142
.254
3.342
1.130
.259
2.599
.382
.702
3.301
1.783
.075*
9.062
-1.927
.054*
9.001
-1.296
.195
1.335
.675
.500
1.188
(4.176)
Asian
-6.347
(3.157)
Hispanic
-13.078
(2.903)
Middle Atlantic
1.564
(3.053)
E. North Central
2.201
(2.863)
W. North Central
.578
(3.266)
South Atlantic
1.232
(2.749)
E. South Central
-.837
(3.393)
W. South Central
-3.412
(2.987)
Mountain
3.560
(3.150)
Pacific
1.145
(2.994)
Population size
.003
(.002)
Population size squared
-5.171E-7
(.000)
Republican
-1.817
(1.402)
Conservative
.971
(1.438)
110
Table 4 (continued)
B
Model
Bible believer
(Std. Error)
t
Sig.
VIF Statistics
-5.331
-3.686
.000***
1.176
(1.446)
Dependent variable: Percentage of correct responses
*significant at .10; **significant at .05; ***significant at .01
R-squared = .497
111
Table 5
Ninety Percent Confidence Intervals for Statistically Significant Coefficients
B
Model
graduate degree
(Std. Error)
29.776
90.0% Confidence Interval for B
Lower Bound
Upper Bound
25.397
34.155
22.622
29.799
14.620
22.589
9.204
15.911
8.352
14.214
6.872
14.231
6.202
11.750
4.973
11.001
5.510
9.055
4.517
9.726
3.338
9.413
-7.713
-2.950
-11.545
-1.149
-16.349
-2.596
(2.659)
bachelor degree
26.210
(2.179)
junior college degree
18.605
(2.420)
Generation Y
12.558
(2.037)
high school diploma
11.283
(1.780)
Very high income
10.551
(2.235)
Generation X
8.976
(1.685)
High income
7.987
(1.830)
Male
7.283
(1.076)
Baby Boomer
7.121
(1.582)
Moderate income
6.376
(1.845)
Bible believer
-5.331
(1.446)
Asian
-6.347
(3.157)
American Indian
-9.472
(4.176)
112
Table 5 (continued)
B
Model
Hispanic
(Std. Error)
-13.078
90.0% Confidence Interval for B
Lower Bound
Upper Bound
-17.858
-8.298
-18.958
-13.687
.000
.006
.000
.000
(2.903)
Black
-16.323
(1.601)
Population size*
.003
(.002)
Population size squared*
-5.171E-7
(.000)
*These are the only continuous variables, all others are dummies.
113
Table 6
Worst-case and Best-case Scenario
Factors
Worst
Best
Constant
Generation
Gender
Income
Education
Ethnicity
Religion
Population size
Total
37.2
0 Traditionalist
0 female
0 very low income
0 less than High school
-16.3 Black
-5.3 Bible believer
.003 (population~1000)
15.6
37.2
12.5 Generation Y
7.2 male
10.5 very high income
29.7 graduate degree
0 White
0 Bible is book of fables
4.5 (population ~3 million)
101.6
The numbers in the table are the truncated values of the mid-point estimates of statistically significant
regression coefficients that fall within a 90% confidence interval. Note that the total for the best-case
scenario is slightly greater than 100.
114
REFERENCES
Bhattacharjee, Y. (2010). Evolution, big bang polls omitted from NSF report. from
http://news.sciencemag.org/scienceinsider/2010/04/evolution-big-bang-pollsomitted.html
Bauer, M. W., Allum, N., & Miller, S. (2007). What can we learn from 25 years of PUS
survey research? Liberating and expanding the agenda. Public Understanding of
Science, 2007, 16, 79-95. doi: 10.1177/0963662506071287
Bennett, D., & Jennings, R. (2011). Successful science communication. New York:
Cambridge University Press.
Cook, J. (2010). New 2009 census data: Income down for the poor, down for the middle
class, up for the rich in the USA. Irregular times. Retrieved from
http://irregulartimes.com/index.php/archives/2010/09/16/new-2009-census-dataincome-down-for-the-poor-down-for-the-middle-class-up-for-the-rich-in-the-usa/
Eckholm, E. (2011, April 3). Pastors hold meetings to revive religious right: Free
conventions aim to boost evangelical political influence. The Sacramento Bee, p.
A12.
General Social Survey (GSS). (1996). Environmental and scientific knowledge around
the world. GSS Cross-National Report No. 16, Chicago, IL: University of
Chicago.
General Social Survey (GSS). (2010). Retrieved from
http://www.norc.org/GSS+Website/ Download/SPSS+Format/
115
Gibson, T. (2004). Culture wars in state education policy: A look at the relative treatment
of evolutionary theory in state science standards. Social Science Quarterly, 85(5),
1129-1149.
Girls RISEnet. (2012). About girls RISENET. Retrieved from http://girlsrisenet.org/about
Gore, A. (2011). Climate of denial: Can science and the truth withstand the merchants of
poison? Rolling Stone, July 7-21. Retrieved from
http://www.rollingstone.com/politics/news/climate-of-denial-20110622
Gross, L. (2006). Scientific illiteracy and the partisan takeover of biology. PLoS Biol
4(5), e167. doi: 10.1371/journal.phbio.0040167
Guterbock, T. (2011). Measurement and operationalization of the “Science in the Service
of Citizens and Consumers” framework. Charlottesville, VA: University of
Virginia.
Institute for Creation Research (ICR). (2012). Home page. Retrieved from
http://www.icr.org/
Intelligent Design and Evolution Awareness Center (IDEA). (2012). Home page.
Retrieved from http://www.ideacenter.org/
Kaufman, L. (2011, February 24). Scientists are cleared of misuse of data. The New York
Times. Retrieved from
http://www.nytimes.com/2011/02/25/science/earth/25noaa.html
116
Kuenzi, J. J. (2008, March 21). Science, technology, engineering, and mathematics
(STEM) education: Background, federal policy, and legislative action. Retrieved
from http://www.fas.org/sgp/crs/misc/RL33434.pdf
Lancaster, L. C., & Stillman, D. (2002). When generations collide. New York:
HarpersCollins Publishers, Inc.
Laugksch, R. (2000). Scientific literacy: a conceptual overview. Science Education, 84,
71-94.
Leiserowitz, A., Smith, N., & Marlon, J. R. (2010). American’s knowledge of climate
change. Yale University. New Haven, CT: Yale Project on Climate Change
Communication. Retrieved from
http://environment.yale.edu/climate/files/ClimateChangeKnowledge2010.pdf
Miller, J. D. (1983). Scientific literacy: A conceptual and empirical overview. Daedalus,
112(2), 29-48.
Miller, J. D. (1998). The measurement of civic scientific literacy. Public Understanding
of Science, 7, 203-223.
Miller, J. D. (2004). Public understanding of, attitudes toward, scientific research: What
we know and what we need to know. Public Understanding of Science, 13, 273294. doi: 10.1177/0963662504044908
Miller, J., Scott, E. C., & Okamoto, S. (2006). Public acceptance of evolution. Science,
313, 765-766.
117
Mooney, C., & Kirshenbaum, S. (2009). Unscientific America: How scientific illiteracy
threatens our future. New York: Basic Books.
Mooney, C. (2005). The Republican war on science. New York: Basic Books.
National Center for Education Statistics (NCES). (2011). National assessment of
education progress (NAEP). Retrieved from
http://nces.ed.gov/nationsreportcard/about/
National Science Foundation (NSF). (2010). Science and engineering indicators: 2010.
Retrieved from http://www.nsf.gov/statistics/seind10/start
National Vaccine Information Center. (2012). Home page. Retrieved from
http://www.nvic.org/
Newsweek (2007, August 13). The truth about denial. Newsweek. Retrieved from
http://www.newsweek.com/2007/08/13/the-truth-about-denial.html
Nisbet, M., & Goidel, R. (2007). Understanding citizen perceptions of science
controversy: bridging the ethnographic survey research divide. Public
Understanding of Science, 16, 421-440. doi: 10.1177/0963662506065558
Pardo, R., & Calvo, R. (2004). The cognitive dimension of public perceptions of science:
methodological issues. Public Understanding of Science, 13, 203-227. doi:
10.1177/0963662504045002
Pitchford, P. (2010). The evolution revolution. UCR Magazine, University of California,
Riverside, 5(1), 8-15.
Pollock, P. (2009). An SPSS companion to political analysis. Washington, DC: CQ Press.
118
Rimer, S. (2011). Study: Many college students not learning to think critically.
McClatchy Washington Bureau. Retrieved from
http://www.mcclatchydc.com/2011/01/18/106949/study-many-college-studentsnot.html
Royal Society. (1985). The public understanding of science. London, England: The Royal
Society.
Shukman, D. (2011). China to overtake US on science in two years. BBC News.
Retrieved from http://www.bbc.co.uk/news/science-environment-12885271
Sills, J. (2010). Climate change and the integrity of science. Science, 328, 6 pp.
Smith, T. (1996). Environmental and scientific knowledge around the world. GSS CrossNational Report No. 16, Chicago, IL: University of Chicago.
Specter, M. (2009). Denialism: How irrational thinking hinders scientific progress,
harms the planet, and threatens our lives. New York: Penguin Press.
Steele, C. (1997). A threat in the air: How stereotypes shape intellectual identity and
performance. American Psychologist, 52(6), 613-629.
Studenmund, A. (2006). Using econometrics: A practical guide (5th ed.). Boston: MA:
Pearson Education, Inc.
Sturgis, P., & Allum, T. (2004). Science in society: Re-evaluating the deficit model of
public attitudes. Public Understanding of Science, 13, 55-74. doi:
10.1177/0963662504042690
119
Sustersic, R. (n.d.). The impacts of religion and education on belief in evolution.
Retrieved from
http://aysps.gsu.edu/econ/files/ECON_SustersicRebecca_Evolution_summer07.p
df
Tucker, J. (2011). Children’s low science proficiency worries leaders. San Francisco
Chronicle. Retrieved from http://www.sfgate.com/cgibin/article.cgi?f=/c/a/2011/01/26/MN381HE38S.DTL
U.S. Census Bureau. (2011). Income inequality. Retrieved from
http://www.census.gov/hhes/www/income/data/historical/inequality/index.html
Vaccination Liberation. (n.d.). Home. Retrieved from http://www.vaclib.org/
White, D. (n.d.). What is the Kyoto protocol? About.com. Retrieved from
http://usliberals.about.com/od/environmentalconcerns/p/KyotoProtocol.htm
Zar, J. (1984). Biostatistical analysis (2nd ed.). Englewood Cliffs, NJ: Prentice Hall, Inc.
Zigerell, L. J. (2010). Science knowledge and biblical literalism. Public Understanding of
Science, 20(10), 1-10. doi: 10.1177/0963662510391723